WO1998019462A1 - Image encoder and image decoder - Google Patents

Image encoder and image decoder Download PDF

Info

Publication number
WO1998019462A1
WO1998019462A1 PCT/JP1997/003976 JP9703976W WO9819462A1 WO 1998019462 A1 WO1998019462 A1 WO 1998019462A1 JP 9703976 W JP9703976 W JP 9703976W WO 9819462 A1 WO9819462 A1 WO 9819462A1
Authority
WO
WIPO (PCT)
Prior art keywords
circuit
block
signal
frame
image
Prior art date
Application number
PCT/JP1997/003976
Other languages
English (en)
French (fr)
Inventor
Noboru Yamaguchi
Toshiaki Watanabe
Takashi Ida
Yoshihiro Kikuchi
Original Assignee
Kabushiki Kaisha Toshiba
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority to BR9706905A priority Critical patent/BR9706905A/pt
Application filed by Kabushiki Kaisha Toshiba filed Critical Kabushiki Kaisha Toshiba
Priority to AU47266/97A priority patent/AU713780B2/en
Priority to CA 2240132 priority patent/CA2240132C/en
Priority to EP97909712A priority patent/EP0873017A4/en
Priority to US09/091,362 priority patent/US6122318A/en
Priority to KR1019980704912A priority patent/KR100286443B1/ko
Publication of WO1998019462A1 publication Critical patent/WO1998019462A1/ja
Priority to NO983043A priority patent/NO983043D0/no
Priority to US09/634,745 priority patent/US6259738B1/en
Priority to US10/703,667 priority patent/US7308031B2/en
Priority to US10/971,284 priority patent/US7215709B2/en
Priority to US10/971,309 priority patent/US7167521B2/en
Priority to US11/683,336 priority patent/US20070147514A1/en

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/50Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
    • H04N19/59Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving spatial sub-sampling or interpolation, e.g. alteration of picture size or resolution
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00Image coding
    • G06T9/20Contour coding, e.g. using detection of edges
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/20Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using video object coding

Definitions

  • the present invention provides a video encoding apparatus and a video decoding apparatus for encoding a video signal (video) with high efficiency, transmitting and storing the video signal, and decoding the video signal.
  • Video decodina apparatus for encoding a video signal (video) with high efficiency, transmitting and storing the video signal, and decoding the video signal.
  • Video decodina apparatus for encoding a video signal (video) with high efficiency, transmitting and storing the video signal, and decoding the video signal.
  • image signals have a huge amount of information, it is common to perform compression-encoding when transmitting or storing them.
  • the image in frame units must be converted to the required number of pixels (for example, MXN pixels (M: number of pixels in the square of water, N: vertical The number of pixels)), the image is divided into blocks, and the divided blocks are subjected to orthogonal transformation to obtain the spatial frequency of the image
  • spatial frequency (spatial frequency) is separated into each frequency component, obtained as a conversion coefficient, and encoded.
  • an alpha-map signa 1 is required. Note that the background alpha-map signal is uniquely obtained from the object alpha-map signal.
  • a binary image encoding method for example, MMR (Modified Modified READ) is used.
  • Coding, etc. or a diagram coding method (chain coding (cha in
  • the contour of the shape is approximated by polygons (po 1 ygons).
  • Os termann a method that is smoothed by spline curves.
  • Ob j ec t -based analys is -synthesis is c cod i ⁇ ⁇ based on the s ource model of f moving ri gid 3 D ob j ec ts s, Signal Proce s s: Image
  • the object is divided into the background and the object, and the object is encoded in order to separate the background and the object. Since an alpha-map signal indicating the shape of the object and its position on the screen is required, the alpha-map of this alpha-map is included together with the image encoding information.
  • the information is also encoded and bitstreamed (it: ream) for transmission and storage.
  • This invention can efficiently encode the information of the alpha map, which is the sub-image information representing the shape of the object and the position in the screen, etc.
  • the purpose of the present invention is to provide an image encoding device and an image decoding device capable of performing the decoding.
  • the alpha image which is information for distinguishing an image into an object area and a background area of the image.
  • the encoding is performed together with the mapping, and the above-described alphabet is encoded by an image encoding device that encodes the encoded data using a relative addres encoding. And refer to the already encoded change pixel (change pixel) as the reference change pixel.
  • a decoding device for decoding a coded bitstream obtained by being coded by the coding device.
  • an image encoding apparatus comprising: means for switching the variable-length encoding table according to the pattern of the alphabet map. Further, the means for switching the variable length encoding table is means for switching according to a pattern near the reference change pixel.
  • the symbol identifying the position of the changing pixel is encoded by using a variable-length encoding table, so that the code amount can be increased.
  • a variable-length encoding table is prepared, and according to the already encoded pattern of the above-mentioned alphabet map. Therefore, the variable length encoding table is switched, and according to the present invention, the Alpha This has the effect of reducing the code amount of the tip.
  • a moving image signal of a plurality of frames (image frames) obtained as time series data is arbitrarily shaped.
  • This is an encoding circuit, and a rectangular area including an object is composed of MXN pixels (M: the number of pixels in the horizontal direction, N: the number of pixels in the vertical direction).
  • the block near the block is used.
  • a motion compensation prediction circuit for generating a motion estimation / compensation alue, and the reproduction value accumulating means. Means for detecting a changed pixel including a reproduction value near a block, and changing a reference changed pixel of the relative address encoding to a pixel value in the block described above. Instead, there is provided a binary image encoding device which can be obtained from a motion compensated prediction signal power.
  • a means for sequentially decoding a rectangular area including an object in accordance with a fixed rule, and a reproduction value in the vicinity of the block Means for storing, image holding means (frame memory) for storing a reproduction signal of a frame (image frame) already encoded, and image holding.
  • a motion compensation prediction circuit that generates a motion compensation prediction value by using the reproduction signal in the means (frame memory), and changes including the reproduction value near the block.
  • Means for detecting a pixel, and the reference change pixel of the relative address encoding is obtained not from the pixel value in the block but from the motion compensation prediction signal.
  • An alpha-map decryption circuit is provided for this purpose.
  • the reference change pixel of the relative address coding obtained from the power and the relative address obtained from the motion compensation prediction signal.
  • a means for sequentially decoding a rectangular area including an object in a regular rule, and a reproduction value in the vicinity of the block Means for storing the reproduced signal of the already encoded frame (image frame), and image storage means (frame memory) for storing the reproduction signal of the already encoded frame (image frame).
  • a motion compensation prediction circuit that generates a motion compensation prediction value using the reproduction signal in the holding means (frame memory), and a reproduction value near the block are also included.
  • the reference change pixel b1 is a block of the image currently being processed in the relative address encoding.
  • the power to be detected from the "current block” and the detection from the "compensated block", which is the block of the previously processed image, are switched for each block.
  • the encoding side also encodes the information for this switching, and the decoding side decodes this information and decodes it.
  • the power to detect the reference change pixel b1 from within the "current block” and the power to detect from the "compensated block” Can be switched on a block-by-block basis, and by doing so, based on the content of the image on a block-by-block basis. Therefore, it becomes possible to perform the optimal processing, and it is possible to perform more efficient encoding.
  • the present invention encodes moving image signals of a plurality of frames obtained as time-series data for each object having an arbitrary shape. Is divided into blocks each consisting of MXN pixels (M: number of pixels in the direction of water square, N: number of pixels in the vertical direction), and this division is also performed.
  • the object represents the shape of an object in an image encoding device that sequentially encodes the inside of the rectangular area according to a fixed rule.
  • the playback signal of the frame including the playback signal in the vicinity of the block and the playback signal of the frame for which encoding has been completed The frame memory that stores the signal and the pixel values in the block are all binary values.
  • a binary image encoding means for encoding a value image and an alpha-map encoding means for encoding the value image are provided.
  • the alpha-map encoding means converts the reproduced image of the block to a reproduced value obtained by replacing all the blocks in the block with one of two values, and a dynamic value.
  • the selected value is either the compensation prediction value or the reproduction value obtained by performing size conversion for each block. Therefore, the encoding of the alpha-map signal can be performed in a high-quality and efficient form while maintaining high-quality image quality and at a high compression rate. And force S.
  • objects are included in order to decode video signals of multiple frames obtained as time-series data for each object of arbitrary shape.
  • the decoding within the rectangular area is performed sequentially according to a fixed rule.
  • the reproduction signal of the frame including the reproduction signal in the vicinity of the block and the reproduction signal of the encoded frame are stored.
  • a motion compensation prediction means for generating a motion compensation prediction value using the reproduced signal of the frame, a means for size-converting a binary image for each block, It is provided with an alpha-map decoding means consisting of a binary image decoding means for decoding a binary image reduced for each lock.
  • the alpha-map decoding means includes: a reproduction value obtained by completely replacing the reproduced image of the block with one of two values in the block; The motion compensation prediction value and the reproduction value obtained by performing size conversion for each block are selected from those. Therefore, high-quality images can be reproduced.
  • an attribute (shape mode) for each block is encoded, and the encoding is performed by an object.
  • Means for dividing, labeling means for assigning a unique label to each attribute (shape mode) for each block, and labeling the label A storage means to be retained for each frame, a determination means for determining a reference frame of a previous frame corresponding to an encoding block of the current frame, and The predicted value is determined at least by the label of the previous frame held in the storage means and the reference block described above.
  • storage means for holding the decoded labels for each frame and A means for determining the reference block of the previous frame corresponding to the decryption block of the current frame, and at least the determination means for storing the reference block.
  • Prediction means for determining a predicted value based on the label of the previous frame and the reference block; and label information of the decoded block, using the predicted value. It is designed to have decoding means for decoding.
  • the analog map is divided into a macro block (for example, when the image is divided into a predetermined multiple pixel configuration such as 16 ⁇ 16 pixels).
  • a unique label is added to the attribute of each block, and this is encoded.
  • the present invention provides a method for encoding an image together with an alpha map, which is information for distinguishing between an object area and a background area of the image.
  • the image encoding device is configured to encode attributes for each block.
  • a storage means for storing the above-mentioned alphabetical map for each frame, and a reference program for a previous frame corresponding to the encoding block of the current frame. Means for deciding the lock, and at least the label having the label of the previous frame held in the storage means is referred to as the alphabet map. Prediction means for determining a predicted value by a reference block; and coding means for coding the label information of the code block using the predicted value. Configuration.
  • a storage means for holding a size conversion rate in a frame unit is provided, and the encoding means includes a frame conversion unit in a frame unit.
  • the size conversion rate is variable, and there is provided means for encoding in accordance with the size conversion rate, and the determining means comprises: a size conversion rate of the current frame; ⁇ Use the size conversion rate of the previous frame obtained from the storage means to obtain the reference frame of the previous frame corresponding to the coding block of the current frame. Be prepared to have a means to determine the impact.
  • storage means for holding a size conversion rate in frame units may be provided, and the encoding means may include a frame size in frame units.
  • the size conversion rate is variable; and means for encoding in accordance with the size conversion rate is provided, and the determining means includes: a size conversion rate of the current frame;
  • the size conversion rate of the previous frame obtained from the storage means is used to calculate the previous frame corresponding to the encoding block of the current frame.
  • Means for determining a reference block of a program, and the predicting means, when there are a plurality of reference blocks, a label of the plurality of reference blocks In particular, provide a means for predicting a large number of labels.
  • storage means for holding a size conversion rate in frame units may be provided, and the encoding means may provide frame size conversion in frame units.
  • the size conversion rate is variable, and is encoded in accordance with the size conversion rate, and the previous frame, the current frame, or both are encoded.
  • the encoding block is encoded using one variable-length encoding table selected from among multiple types according to the size conversion rate of Means for determining the current frame using the size conversion rate of the current frame and the size conversion rate of the previous frame obtained from the storage means.
  • the decoding device that reproduces the attributes of each block of the alpha map is reproduced, and the decoded label or the alpha map is used.
  • Storage means for storing a frame for each frame, and determination means for determining a reference frame of a previous frame corresponding to a decoding block of the current frame. At least, the forecast value is determined by the label or alphabet of the previous frame held in the storage means and the reference block.
  • a decoding means for decoding the label information of the decoding block by using the above-mentioned predicted value. It is characterized by
  • the size conversion rate of the frame is variable on a frame basis, a means for decoding the size conversion rate information, and a method for decoding the size conversion rate information.
  • a storage means for holding wherein the determining means includes a size conversion rate of a current frame and a size conversion rate of a previous frame read out from the storage means. It has a function to determine the reference block of the previous frame corresponding to the decryption block of the current frame by using .
  • the size conversion rate of the frame is variable in frame units, a means for decoding the size conversion rate information, and a method for decoding the size conversion rate information.
  • a storage means for holding wherein the determining means includes a size conversion rate of a current frame and a size conversion rate of a previous frame read out from the storage means.
  • a function to determine the reference frame of the previous frame corresponding to the decryption block of the current frame using the current frame is provided.
  • the feature is that, among the labels of the multiple reference blocks, the number, the number of labels, and the label are the predicted values. .
  • An image coding apparatus for sequentially coding the rectangular area according to a fixed rule for each frame, and includes an object in the frame.
  • Setting means for setting an area represented by a multiple of the block size, and dividing the area set by the setting means into blocks. Dividing means and a moving vector necessary for predicting the movement in the divided block
  • an image coding apparatus having means for predicting (motion vector) coding
  • the position of the region in the reference frame in the frame is determined.
  • the memory holding the first position vector to be represented and the second position vector representing the position of the reference frame in the frame within the frame are denoted by a code.
  • Encoding means for encoding a moving vector memory for holding a moving vector of a reproduced block in the vicinity of a block to be encoded, and a moving vector memory for storing the moving vector.
  • a means is provided for predicting the motion vector of the block to be encoded by using the motion vector stored in the vector memory. Ere, If the motion vector used in the prediction means does not exist in the motion vector memory, the default motion vector is used. Is the predicted value, and the default dynamic vector is the difference vector between the first and second position vectors, and zero. It is characterized in that it can be used by switching the bottle.
  • the moving image signal of a plurality of frames obtained as time series data is decoded for each object of an arbitrary shape, and the MXN including the object is decoded.
  • the rectangular area is sequentially and sequentially decoded according to a fixed rule. Therefore, within the frame, the image decoding that reproduces the area represented by a multiple of the block size including the object for each block Means for decoding a predicted encoded motion vector necessary for performing motion compensation prediction in the block, and a reference frame.
  • a means for decoding the predicted encoded motion vector necessary for motion compensation prediction A memory for holding a first position vector representing a position in the frame of an area in the reference frame; and a memory for holding a first position vector in the frame of the area in the reference frame. Means for decoding the second position vector representing the position at, and retaining the motion vector of the corrected block near the block to be decoded Using the moving vector memory and the moving vector held in the moving vector memory, the moving of the block to be decoded is performed. Prediction method to predict the vector If the motion vector to be used in the prediction means does not exist in the motion vector memory, the default motion vector is used. This vector is the predicted value, and the default motion vector is the difference vector between the first position vector and the second position vector. It is characterized in that it is either zero vector or zero vector.
  • FIG. 1 is a schematic block diagram of an encoding device to which the present invention is applied.
  • FIG. 2 is a schematic block diagram of a decoding device corresponding to the coding device of FIG.
  • FIG. 3 is a block diagram of an alpha-map encoding circuit of an encoding device to which the present invention is applied.
  • FIG. 4 is a block diagram showing a configuration of an alpha-map decoding circuit applied to a decoding device corresponding to the coding device of FIG.
  • FIG. 5 is a block diagram of an encoding circuit according to the first embodiment of the present invention.
  • FIG. 6 is a block diagram of a decoding circuit corresponding to the coding circuit of FIG.
  • FIGS. 7A and 7B are diagrams showing the relationship between changed pixels when encoding is performed in block units, and diagrams showing a reference area for detecting b1 (protocol). A diagram showing the relationship between reference pixels and the change pixels of the code base coding).
  • FIG. 8 is a flowchart showing a case where the MMR is encoded on a block basis.
  • FIG. 9 is a diagram for explaining the operation of the encoding circuit of FIG. 5, and is a diagram showing an example of the state of the periphery of the changing pixel b1 (FIG. 10 is a diagram of FIG. 6).
  • FIG. 5 is a diagram for explaining the operation of the decoding circuit, and is a diagram showing an example of a reference pixel.
  • Figure 11 is a diagram for explaining how to determine the context number.
  • FIG. 12 is a block diagram of an encoding circuit according to a second embodiment of the present invention.
  • FIG. 13 is a block diagram of a decoding circuit 'corresponding to the coding circuit of FIG.
  • FIG. 14 is a block diagram of an encoding circuit according to a third embodiment of the present invention.
  • FIG. 15 is a block diagram of a decoding circuit corresponding to the coding circuit of FIG.
  • FIGS. 16A and 16B are diagrams for explaining the third embodiment of the present invention, in which the transformation pixels in the inter-frame encoding are illustrated. Diagram explaining the detection circuit.
  • FIGS. 17A and 17B are diagrams illustrating switching in the scan direction.
  • Fig. 18 shows how the screen of the Alpha Map used in the present invention is divided into macroblock (MB) units with a predetermined multiple pixel configuration.
  • FIG. 19 is a diagram showing the alpha-female according to the fourth embodiment of the present invention. Block of coding circuit
  • FIG. 20 is a block diagram of an alpha-map decoding circuit corresponding to the coding circuit of FIG.
  • Figure 21 is a diagram illustrating Markov model encoding.
  • Figure 22A is a binary image that switches between multiple binary surface image encoding methods. Block diagram of image encoding circuit.
  • Fig. 22B is a block diagram of a binary image decoding circuit that uses multiple binary surface coding schemes.
  • Figures 23A and 23B are diagrams illustrating the bilinear interior used in the size transformation (reduction / enlargement) processing.
  • Figures 24A and 24B are diagrams for explaining the smoothing processing (smoothing processing).
  • FIG. 25 is a diagram for explaining another example of a smoothing filter (smoothing filter) used in the present invention.
  • FIGS. 26A and 26B are diagrams for explaining an example of a process of double-expanding both horizontally and vertically by bilinear interpolation.
  • Fig. 27A to 27D is a diagram explaining the inner pixel position and the use range of the reference pixel in the enlargement processing applied in the present invention.
  • FIG. 4 is a diagram illustrating an example of reference pixel addition processing in the enlargement processing applied in the present invention.
  • FIG. 29 is a diagram for explaining another example of the size conversion processing for each block.
  • Figure 30 illustrates an example of the reduction process that reduces a block (macro block) to the size of "1/2" vertically and horizontally.
  • FIG. 31 is a diagram for explaining an example of a method for obtaining a pixel value of a reduced block.
  • FIG. 32 is a diagram for explaining an example of reduction processing by pixel thinning.
  • FIG. 33 is a diagram for explaining an example of the enlargement processing used in the present invention.
  • Figures 34A and 34B illustrate the processing contents that are doubled both horizontally and vertically by the bilinear internal processing used in the present invention.
  • FIG. 35 is a diagram showing an example of a combination of the size conversion in units of frames and the size conversion in units of small areas in the fifth embodiment of the present invention. Block diagram of the encoding circuit.
  • FIG. 36 is a block diagram of an alpha-map decoding circuit corresponding to the alpha-map coding circuit of FIG. 35.
  • FIG. 37 is a diagram showing an example of the resolution for each frame.
  • FIG. 38 is a block diagram of the encoding device when the frame memory required for the encoding device of FIG. 35 is explicitly included and represented.
  • FIG. 39 is a block diagram of the decryption apparatus when the frame memory required for the decryption apparatus of FIG. 36 is explicitly included and represented.
  • FIG. 40 is a block diagram of another encoding device in a case where the frame memory required for the encoding device of FIG. 35 is explicitly included and represented.
  • FIG. 41 is a block diagram of another decryption device when the frame memory necessary for the decryption device of FIG. 36 is explicitly included and represented.
  • Figure 42A is a block diagram of the frame memory used in the encoding system of the present invention.
  • Fig. 42B is a block diagram of the frame memory used in the decoding device of the present invention.
  • FIGS. 43A and 43B are diagrams for explaining the technology related to the sixth embodiment of the present invention.
  • FIGS. 44A and 44B are diagrams for explaining the technology related to the sixth embodiment of the present invention.
  • Figures 45A and 45B show the frame images Fn-1 and Fn at time n-1 and time n, respectively, of the present invention and their respective frames Fn.
  • -Diagram for explaining an example of the mode information MD of each macro block of the coding area CA in -1 and Fn.
  • FIGS. 46A and 46B are diagrams for explaining the technology related to the sixth embodiment of the present invention.
  • FIG. 47A is a diagram for describing an encoding device related to the sixth embodiment of the present invention.
  • FIG. 47B is a diagram for describing a decoding device related to the sixth embodiment of the present invention.
  • FIG. 48 is a block diagram of an encoding circuit according to a sixth embodiment of the present invention.
  • Figures 49A and 49B show time n — 1 and time n.
  • Figures 50A and 50B are diagrams for explaining the change of the coding area and the change of the position of the corresponding block in the mode Iff information.
  • 51 1 is a moving picture coding device that uses
  • FIG. 52A A diagram for explaining the encoded data configuration in Jv_. Fig. 52A to Toshi Fig. 52D explains how to deal with undetermined labels that occur when the target area occupied by the coding area is a part of the frame. Figure for.
  • Fig. 53 is a diagram for explaining an example of the reduction process used in the present invention.
  • FIG. 54 is a diagram for explaining the reduction processing used in the present invention.
  • Fig. 55 Fig. 5 is a diagram for explaining the label prediction method used in the present invention.
  • 5 6 is a diagram for explaining the process of restoring a frame from the reduced frame used in the present invention by the enlargement process.
  • Fig. 57 is a block diagram of the encoding device of the present invention using labels for prediction.
  • FIG. 58 is a flowchart showing an example of an encoding processing unit used in the encoding device of the present invention.
  • FIG. 59 is a flowchart showing another example of the encoding processing procedure used in the encoding device of the present invention.
  • FIG. 60 is a view for explaining the order of the bit strings output from the alpha-map encoding circuit of the present invention.
  • FIG. 61 is a block diagram showing an example of a system configuration according to a sixth embodiment of the present invention.
  • FIG. 62 is a flowchart for explaining a processing procedure according to the sixth embodiment of the present invention.
  • FIG. 63 is a flowchart for explaining the processing procedure in the sixth embodiment of the present invention.
  • FIG. 64 is a diagram for explaining an example of predictive coding of a motion vector used in the present invention.
  • Figures 65A and 65B illustrate the inconvenience of predicting the motion vector when the position of the object within the frame is highly variable. Figure for simplicity.
  • FIGS. 66A and 66B are block diagrams of the MV encoding circuit and its peripheral circuits in the system of the present invention.
  • Fig. 67A is a block diagram of the MV regeneration circuit in the system of the present invention.
  • FIG. 67B is a block diagram of a peripheral circuit of the MV reproduction circuit in the system of the present invention. Best form to carry out the invention
  • FIG. 1 shows an image encoding apparatus to which a method of encoding an image by dividing a screen into a background and an object and encoding the divided image is applied.
  • This image coding apparatus includes a difference circuit (subtracacter) 10 and a motion compensation prediction circuit.
  • the alpha-map encoding circuit 20 encodes the input alpha-map signal and converts the encoded signal to alpha-format. Multiplexing circuit as the top signal
  • the Alpha-map encoding circuit 20 encodes an Alpha-map signal input in units of blocks.
  • the given reduction rate (conversion) conversion
  • ratio (ratio) the resolution of the alpha map
  • the difference circuit 10 is provided with a motion compensation prediction signal (motion) supplied from the motion compensation prediction circuit 11.
  • difference signal difference signal
  • orthogonal transformation circuit 12 converts the difference signal supplied from the difference circuit 10 into an alpha map. According to the information, they are converted to orthogonal transform coefficients (or orthogonal transform coefficients) and output.
  • the quantization circuit 13 is a circuit for quantizing the orthogonal transform coefficients obtained by the orthogonal transformation circuit 12, and the variable length coding circuit 14 is a quantization circuit. This is a circuit that encodes the output of 13 and outputs it.
  • the multiplexing circuit 18 is composed of a signal encoded by the variable-length encoding circuit 14 and an alphabet. The output signal is multiplexed with the side information such as motion vector information and output as a bit stream.
  • the inverse quantization circuit 15 has a function of inversely quantizing the output of the quantization circuit 13, and the inverse orthogonal transform circuit 16 has an output of the inverse quantization circuit 15 It has a function to perform inverse orthogonal transform based on the hammap signal.
  • the addition circuit 17 adds the output of the inverse orthogonal transform circuit 16 and a prediction signal (motion compensation prediction signal) provided from the motion compensation prediction circuit 11 to add. Output to the motion compensation prediction circuit 11.
  • the motion compensation prediction circuit 11 has a frame memory, and is based on a local decoding signal provided from the alpha-map coding circuit 20. It has a function to operate and accumulate the signal of the object area and the signal of the background area. Further, the motion compensation prediction circuit 11 predicts the motion compensation value from the stored image area of the image area, outputs the predicted value as the predicted value, and outputs the accumulated background. Motion compensation value from the image of the area 3 ⁇ 4r Predicted and predicted value
  • An image signal and an alpha-map signal of the image signal are input to the encoding device.
  • Each of these signals has a predetermined pixel size (for example, MXN pixels (M: number of pixels in the horizontal direction, N: number of pixels in the vertical direction)) for each frame.
  • This division processing is performed by using a memory (shown in the drawing) that stores an image signal for each frame according to a known technique. This is done by designating the address and reading out the image signal in block units.
  • the image signal for each block obtained by this division processing is supplied to the differential circuit 10 via the signal line 1 in the order of the block position.
  • the difference signal between this input (image signal) and the prediction signal (the output of the motion compensation prediction signal from the op- tim prediction circuit 11) is calculated. , Supplied to the orthogonal transformation circuit 12.
  • the orthogonal transformation circuit 12 converts the supplied difference signal into an alpha-map signal supplied from an alpha-map encoding circuit 20 via a signal line 4. After the orthogonal transform coefficients are converted into orthogonal transform coefficients according to the information of the top, the orthogonal transform coefficients are supplied to a quantization circuit 13 where they are quantized. The transform coefficients obtained by quantization in the quantization circuit 13 are encoded in the variable-length encoding circuit 14, and are also encoded by the inverse quantization circuit 1. Supplied to 5.
  • the transform coefficient supplied to the inverse quantization circuit 15 is inversely quantized here and then inversely transformed in the inverse orthogonal transformation circuit 16.
  • the inverse transform coefficient is added to the motion compensation prediction value supplied from the motion compensation prediction circuit 11 in the addition circuit 17 and output as a local decoded image. Is input to the motion compensation prediction circuit 11 again.
  • the local decoded image output from the adding circuit 17 is stored in a frame memory in the motion compensation prediction circuit 11.
  • the motion compensation prediction circuit 11 The timing of the processing of the block in the object region based on the local decoding signal provided from the encoding circuit 20 is "Object".
  • the motion compensation predicted value of the subject is output to the difference circuit 10 at other times, and the motion compensation predicted value of the background portion is output to the difference circuit 10 at other timings.
  • the motion compensation prediction circuit 11 is configured to convert the local decoding signal of the alphabet signal to the image signal of the block corresponding to the block of the object. Is input to the differential circuit 10, or whether the image signal of the block-corresponding portion of the background portion is input to the differential circuit 10? It is known that, during the input period of the image signal of the block corresponding to the object, the motion compensation prediction signal of the object and the background portion are used. If the block-corresponding part is during the image signal input period, the background motion compensation prediction signal is given to the difference circuit 10.
  • the difference circuit 10 calculates the difference between the input image signal and the prediction signal corresponding to the area of the image, and as a result, the input image If the image is in an area corresponding to the object, the difference signal between the predicted value at the corresponding position of the object and the input image, and the input image is the area in the background. If so, a difference signal between the predicted value corresponding to the background position and the input image is calculated and supplied to the orthogonal transformation circuit 12.
  • the orthogonal transformation circuit 12 converts the supplied difference signal into information according to the information of the alpha map supplied through the signal line 4, After being transformed into orthogonal transform coefficients, it is supplied to a quantization circuit 13. The orthogonal transform coefficients are quantized by the quantization circuit 13.
  • the transform coefficients quantized by the quantization circuit 13 are encoded by the variable-length encoding circuit 14 and supplied to the inverse quantization circuit 15. It is done.
  • the transform coefficient supplied to the inverse quantization circuit 15 is inversely quantized here, then inversely transformed in the inverse orthogonal transformation circuit 16 and supplied to the addition circuit 17.
  • the predicted value supplied to the addition circuit 17 from the motion compensation prediction circuit 11 is calculated.
  • the signal of the local decoded image which is the output of the adding circuit 17, is supplied to the motion compensation prediction circuit 11.
  • the motion compensation prediction circuit 11 is a block of the object from the local recovery signal of the alpha-map signal to the current circuit 17. It knows whether the signal corresponding to the lock is output, or the signal corresponding to the block in the background is output and re-read, and the result is obtained. If a signal corresponding to the object block is being output, the frame memory for the object and the background block are output. If the corresponding signal is being output, it operates as much as possible for the background memory and stores the signal in the corresponding memory. As a result, only the object image is stored in the frame memory for objects, and only the background image is displayed in the background memory. Is obtained. Therefore, the motion compensation prediction circuit 11 can obtain the predicted value of the object image by using the object image. In addition, the predicted value of the background image can be obtained by using the image of the background portion.
  • the input alpha-map is encoded, and the encoded alpha-map is encoded.
  • the LF map signal is supplied to the multiplex circuit 18 via the signal line 3 and is relayed.
  • the conversion coefficient output from the variable length coding circuit 14 is supplied to the multiplexing circuit 18 via the line 4.
  • the multiplexing circuit 18 is supplied, and the encoded values of the alpha-map signals and the conversion coefficients are converted to side information such as motion vector information.
  • the signal is output via a signal line 5 to form a code-activated stream as the final output of the image coding device.
  • the encoder performs motion compensation prediction based on the object and background images in order to obtain the error signal of the image. In other words, whether the current block position of the image being processed according to the alpha map is the position of the object region or the position of the background region is determined. If the current block position of the image being processed is the position of the object area, the predicted value obtained from the image for the object is used. If it is a position, the difference is calculated using the predicted value obtained from the background image.
  • FIG. 2 is a block diagram of a decoding apparatus to which the present invention is applied.
  • the decoder includes a demultiplexer 30, a variable length decoder 31, a dequantizer, and a dequantizer.
  • inverse orthogonal transform circuit (adder) 3 4 motion compensation circuit (motion)
  • the separation circuit 30 composed of an N -A / N map is an input encoding circuit. This is a circuit for demultiplexing the stream to obtain an alpha-map signal and an image-encoded signal, etc.
  • the decoding circuit 40 is a circuit for decoding the encoded alpha-map signal separated by the demultiplexing circuit 30.
  • the variable length decoding circuit 31 is a circuit for decoding the coded image signal separated by the separation circuit 30, and the inverse quantization circuit 32 is the decoding circuit for decoding the coded image signal. It has a function of dequantizing the obtained image signal and returning it to the original coefficient, and the inverse orthogonal transform circuit 33 performs an inverse orthogonal transform of this coefficient according to the alpha map. It has a function of returning to the prediction error signal.
  • the motion compensation value from the motion compensation circuit 35 is added to the measurement error signal, and the result is output as a decoded video signal (decoded video signal). This reproduced image signal is the final output of the decoding device.
  • the motion compensation circuit 35 stores the reproduced image signal output from the addition circuit 34 in the frame memory according to the alphabet map. In addition to obtaining the object image and the background image, the motion compensation signal of the object and the motion of the background are obtained from the accumulated and obtained images. Obtain compensation signal.
  • the coded bit stream is supplied to the demultiplexing circuit 3 via the line 7 and the demultiplexing is performed.
  • the code for the alpha-map signal and the variable-length code for the image signal are separated. Can be separated.
  • the code related to the alpha-map signal is supplied to the alpha-map decoding circuit 40 via a signal line 8, and the variable length of the image signal is changed.
  • the code is supplied to a variable length decoding circuit 31.
  • the code related to the alpha-map signal is reproduced by the alpha-map decoding circuit 40 as the alpha-map signal, and the signal line 9 is transmitted.
  • the signal is output to the inverse orthogonal transform circuit 33 and the motion compensation circuit 35 via the inverter.
  • variable-length decoding circuit 31 decodes the code supplied from the demultiplexing circuit 30, and the decoded transform coefficient is inverted.
  • the signal is supplied to a quantization circuit 32, where it is dequantized.
  • the inversely quantized transform coefficient is inversely transformed by an inverse orthogonal transform circuit 33 in accordance with an alpha-map supplied via a line 9, and is added to the adder circuit 3. Supplied to 4.
  • the addition circuit 34 adds the signal subjected to the inverse orthogonal transform from the inverse orthogonal transform circuit 33 to the motion compensation signal supplied from the motion compensation circuit 35, and reproduces the signal. Obtain an image.
  • the present invention is a component of the alpha-map encoding circuit 20 which is a component of the encoding device shown in FIG. 1 and a component of the decoding device shown in FIG.
  • the present invention relates to the alpha-map decoding circuit 40, and a specific embodiment will be described below.
  • a symbol that identifies the location of the changing pixel is encoded using a variable-length encoding table, and the symbol that frequently appears is encoded in a symbol that frequently appears.
  • the already-encoded / decoded pixels of the alpha map By changing the length of the variable length coding table (VLC table) adaptively, the amount of coding can be further reduced.
  • a symbol for specifying the position of a changing pixel is encoded using a variable-length encoding table, and the symbol is identified.
  • Alpha map information encoded in.
  • the feature is that the variable length encoding table is switched according to the turn.
  • the pixels of the file map that have already been encoded / decoded can be further reduced by adaptively switching the variable length coding table (VLC tape) according to the turn.
  • VLC tape variable length coding table
  • the alpha-map encoding circuit 20 is a resolution conversion circuit (fezj ⁇ LEI path (down-sampling circuit)) 2 1, (enlargement circuit (up-sampling circuit) 2) 3, a binary image encoder 22 and a multiplexer 24 are provided.
  • the resolution conversion circuit 21 is a conversion circuit for lowering the resolution (scaling-down), and is provided with a given reduction, j, and enlargement ratio (conversion ratio).
  • the resolution conversion circuit 23 is a conversion circuit for increasing the resolution (scaling-up). It has a function to enlarge the alphabet map according to the obtained reduction / enlargement ratio.
  • the resolution conversion circuit 23 is provided in order to restore the original image having the reduced resolution converted by the resolution conversion circuit 21 to the original size, and this resolution conversion circuit 23 is provided.
  • Road 23 shows the original
  • the alpha map returned to the size is supplied to the orthogonal transform circuit 12 and the inverse orthogonal transform circuit 16 via the signal line 4. It becomes a local decoding signal.
  • the binary image encoding circuit 22 converts the resolution-reduced and converted alpha-map signal (down-sampled alpha-map signal) output from the resolution conversion circuit 21 into binary.
  • the multiplexing circuit 24 has a function of outputting an image encoded and outputs the binary image encoded output and information of the reduction / enlargement rate given via the signal line 6. And are multiplexed and output.
  • the alpha-map input via the alpha-map signal input line 2 is used.
  • the map signal is encoded after being reduced by the resolution reduction circuit 21 at the specified reduction / enlargement ratio, and the encoded alpha-map signal is encoded. Via the signal line 3 and the encoded alpha-map signal.
  • the desired size conversion rate (stre) is input to the alpha-map encoding circuit 20 via the signal line 6.
  • the size conversion rate supplied via signal line 6 is equal to the resolution
  • the number of encoded bits (the number of encoded bits) supplied to the degree conversion circuits 21 and 23 and the binary image encoding circuit 22 Is controlled by the force S.
  • the size conversion rate code (setting information signal) supplied via the signal line 6 is encoded by the multiplexing circuit 24.
  • the multiplexed signal is multiplexed with the converted alpha-map signal, output via the signal line 3, and used as an encoded signal of the alpha-map as an image encoding device. This is supplied to the multiplexing circuit 18 which is the final output stage.
  • the alpha-map decoding circuit 40 is composed of a binary image decoding circuit 41, a resolution conversion circuit 42, and a demultiplexing circuit 43. It is composed of The demultiplexing circuit 43 is separated by the demultiplexing circuit 30 in the image decoding apparatus shown in FIG. 2 and input to the alpha-map decoding circuit 40. Separates the code of the alpha-map signal into the code of the alpha-map signal and the code of the size conversion rate (the setting information signal of the size conversion rate) (demultiplex).
  • the binary image decoding circuit 41 is a circuit that separates the sign of the alpha-map signal from the separation circuit 43 and provides the code. This is a circuit for returning to a binary image in accordance with the sign of the conversion rate, and the resolution conversion circuit 42 is provided by separating this binary image from the separation circuit 43. Resolution according to the sign of the size conversion rate Large conversion and output.
  • the code supplied to the alpha-map decoding circuit 40 via the signal line 8 is converted into the alpha-map format by the separation circuit 43.
  • the signal is separated into the sign of the Tup signal and the sign of the size conversion rate, and output via the signal line 44 and the signal line 45, respectively.
  • the code of the alpha-map signal supplied via the signal line 44 and the signal supplied via the signal line 45 are provided. From the code of the size conversion rate, the reduced analog map signal is reproduced and supplied to the resolution conversion circuit 42 via the signal line 46. In the resolution conversion circuit 42, the reduced alpha-map signal is converted from the code of the size conversion rate supplied via the signal line 45 to the original size. After that, the signal is output via the signal line 9.
  • FIG. 5 shows the alpha-map encoding circuit 20 of FIG. 1 in more detail of the 2D direct image encoding circuit 22 of FIG. 3! /
  • the alpha map signal 51 is input to the a1 detection circuit 52 and the memory 53 that holds the encoded alpha map.
  • the position of the change pixel a1 is detected, and the position signal 54 is output.
  • FIG. 7A shows a block unit (for example, a block unit of an MXN pixel configuration (M: the number of pixels in the horizontal direction, N: the number of pixels in the vertical direction)).
  • Encode Alphamap signal FIG. 4 is a diagram showing the relationship of changing pixels when changing.
  • FIG. 7B shows the reference change pixel.
  • (reference changing ixel) b1 is a diagram showing a reference area (reference area) for detection.
  • the changing pixels may be simplified and encoded as follows.
  • the following processing may be performed by switching the scan order or by applying it to a reduced block.
  • the coding of the simplified changing pixel is performed as follows.
  • r _b 1 abs— bl— (aO— line— 1) * WIDTH
  • * means multiplication
  • (int) (X) means truncation of X below the decimal point.
  • WIDTH indicates the number of pixels in the block in the water square direction. Change pixel phase For a certain address, "r--al--r--bl", a reproduction value can be obtained by encoding the value of "r--al--r-1a0". In this way, the position of the transformation pixel a1 is detected.
  • the information of the position 54 detected by the a1 detection circuit 52 is sent to a shape mode determination circuit ⁇ 5.
  • the memory 53 sends the position signal 56 of the reference change pixel b1 to the mode determination circuit 55.
  • the mode is determined by the algorithm shown in FIG. 8, and the mode is encoded by a symbol.
  • the signal is sent to the encoding circuit 58 as 57.
  • the position of the starting-point change pixel is initialized (S 1), and the pixel value at the initial position (the upper left pixel of the block) is encoded in one bit (S 2). Then, at the initial position, the reference change pixel b1 is detected (S3).
  • the vertical mode cannot be used because the change pixel does not exist in the reference area, so that the vertical mode cannot be used. No ,.
  • the state of the mode is set to "TRUE”. If b1 is detected, the vertical mode can be used.
  • the changed pixel a1 is detected (S5), and it is determined whether or not the changed pixel a1 is detected (S6). If 1 is not detected, since there is no changed pixel, the encoding process end code (EOMB; End of MB) indicating the end of encoding is marked. (S7).
  • EOMB End of MB
  • step S6 if a changed pixel a1 is detected, the state of the vertical pass mode (Vertical pass mode) is determined (S6). 8). Here, vertical. If the mode status is "TRUE”, the vertical pass mode is encoded (S16), and the vertical pass mode is set. If the mode status is "FALSE”, b1 is detected (S9).
  • step (S10) it is determined whether or not b1 has been detected (S10) . If the reference change pixel b1 has not been detected, the horizontal mode is switched off. Proceeding to step (S 13), when the reference change pixel b 1 is detected, the absolute value of “r—a 1 —r—b 1” is set to the threshold value (VTH ) Is determined (S 11), and as a result, if it is below the threshold (VTH), the vertical mode step (S 11 Proceed to 1 2), and if it is greater than the threshold (VTH), proceed to the horizontal mode step (S 13).
  • VTH threshold value
  • the vertical pass mode status is set to "TRUE” (S15), and the vertical pass mode is set. Steps in mode (S 1 Proceed to 6), and when the vertical nos mode step (S16) ends, set the vertical pass mode status to "FALSE" (false).
  • the notation 59 around the encoded reference change pixel b1 is obtained from the memory 53. Is sent to the table determination circuit 60. In the table determination circuit 60, one of the plurality of variable-length encoded tables is selected and output.
  • the reference change pixel Since the same edge often extends linearly below b 1, among pixels X 1, X 2, and X 3, X 1 has a 1 power S high probability S high among pixels X 1, X 2, and X 3. Ray.
  • the upper part of the reference change pixel b 1 is such a pixel.
  • the value obtained by converting this binary number to a decimal number is called a context number. For example, corresponding to each context number,
  • V L 1 represents b
  • V L 2 is b-2
  • V R 1 is b
  • the encoding circuit 58 determines the code 62 by using the selected table 11 sent from the table determination circuit 60, and determines the code.
  • the code 62 is output.
  • FIG. 6 is a diagram showing an example of the configuration of the decryption circuit 40 as a component of the decryption device shown in FIG.
  • a block diagram showing the binary image decoding circuit 41 as a component of the block diagram in more detail is shown. This decodes the code 62 generated in the embodiment of FIG.
  • the code 62 is input to the decoding circuit 63.
  • the memory 64 retains the previously decoded alphabet map, and the area around the reference change pixel b1.
  • the turn 65 is sent to the table decision circuit 66.
  • the table determination circuit 63 one of a plurality of variable-length encoded TFs is transmitted to the decoding circuit 63 as a table 67 in which one selected. It is.
  • the algorithm for determining the table is the same as the table determination circuit 70 in FIG.
  • the symbol 68 is decoded and sent to the a1 reproduction circuit 69.
  • a 1 In the reproduction circuit 69 the position of a 1 is obtained from the position 70 of b 1 sent from the symbol 68 and the memory 64, and a 1 Play the previous fanmap 71. Replayed Alpha Map
  • FIG. 12 shows the actual occurrence of a symbol.
  • a second embodiment in which the table is dynamically modified depending on the frequency is described below.
  • FIG. 12 shows that the number of variable-length encoding tables depends on the frequency of the symbols actually generated.
  • FIG. 5 shows an embodiment in which the table is dynamically modified.
  • the encoding device of the first embodiment in FIG. 5 is provided with a counter 72, and a header and a circuit generating circuit (Hoffman table forming circuit) 73. It is.
  • the counter 72 has a mode determination circuit 55, a symbol 57 from a power source, and a table determination circuit 60, a context number from a power source 60. 7 4 is input.
  • the counter 72 holds the number of occurrences of each symbol for each context number. After a certain period of time has passed from this holding power, the number of occurrences of each symbol for each context number is 75, and the number of generations of the symbol is no. It is sent to circuit 73.
  • the ⁇ - knob encoding is used (Fujita “Basic Information Theory” (Shokodo)). 52-53, 1987 ) Generates an encoded table 76.
  • the table 76 is sent to the table determination circuit 60, and is replaced with the table power S table 76 of the corresponding context number. It is. The generation and replacement of this artifact sample is performed for all context numbers.
  • FIG. 13 shows a decoding circuit for decoding the code generated by the coding circuit of FIG. 12.
  • the decoding device according to the second embodiment shown in FIG. 13 is also replaced by the decoder 77 according to the embodiment shown in FIG. This is a configuration in which the road 78 is added.
  • the operations of the power counter 77 and the noise mountable generation circuit 78 are the same as those of 2.
  • the symbol for specifying the position of the changed pixel is encoded by using the variable length encoding table.
  • a plurality of types of variable length encoding tables are used and already used.
  • the variable-length coding table is switched according to the pattern of the coded file map. According to the present invention as described above, it is possible to obtain an effect that the code amount of the alphabet map can be reduced.
  • the reference change pixel of the relative address coding is referred to as MX. It can be obtained from the motion compensation prediction signal, not from the pixel value in the block of N pixels (M: number of pixels in the direction of water square, N: number of pixels in the vertical direction).
  • M number of pixels in the direction of water square
  • N number of pixels in the vertical direction.
  • FIG. 14 is a block diagram for explaining an alpha-map encoding circuit according to the third embodiment.
  • FIG. 15 is a block diagram for explaining an alpha-map decoding circuit according to a third embodiment.
  • FIG. 14 Using FIG. 14, FIG. 15 and FIG. 16A, 16B, the alpha-map encoding circuit 20 and alpha-map of the present invention are used.
  • the decryption circuit 40 will be described.
  • the file map encoding circuit 20 is a resolution conversion circuit (reduction processing circuit i down- sampling c ircuit)) 2 1, resolution conversion circuit (up-sampling c ircuit)) 2 3, binary video encoding c ircuit, for exam le, block -based MMR encoder) 2 2, multiplexer 24, and motion compensation prediction circuit
  • the resolution conversion circuit 21 is a conversion circuit for resolution down conversion (down-sampling), and has a given size conversion ratio (size convers ion ratio).
  • Configuration The alpha-map signal is encoded at a reduction rate according to the information signal
  • the resolution conversion circuit 23 is a conversion circuit for resolution enlargement conversion (up-sarapling).
  • it has a function to encode an alpha map at a magnification rate according to a given magnification rate.
  • the resolution conversion circuit 23 is provided in order to return the Alpha Map whose resolution has been reduced by the resolution conversion circuit 21 to the original size, and this resolution conversion circuit 23 is provided.
  • the alpha map returned to its original size by 23 is transmitted via the signal line 4 to the orthogonal transformation circuit 12 and the inverse orthogonal transformation circuit 16 in FIG. This is the alpha-map local decoding signal that is input to.
  • the binary image encoding circuit 22 encodes the resolution-converted (down-sampled) alpha-map signal output from the resolution conversion circuit 21 into a binary image.
  • the resolution conversion circuit 26 for reduction processing is supplied from the resolution conversion circuit 26 for reduction processing to the signal line 82.
  • the encoding is performed using the motion compensation prediction signal of the firmware.
  • the multiplexing circuit 24 multiplexes the binary image encoded output and the information on the enlargement ratio and outputs the result.
  • the configuration of the encoding circuit in the third embodiment is different from the encoding circuit shown in FIG. 1 in that a motion compensation prediction circuit 25 and a resolution conversion circuit 26 for reduction processing are provided.
  • the motion compensation prediction circuit 25 is provided with a frame memory for storing the reproduced image of the previously encoded frame.
  • the reproduced signal (decoded) supplied from the expansion circuit 23 signa 1) Force S
  • the motion compensation prediction circuit 25 also has a motion vector signal.
  • a motion compensation prediction signal is generated according to the motion vector signal, and a resolution conversion circuit for reduction processing is provided via a signal line 81. 26.
  • a resolution conversion circuit 26 for reduction processing supplies, via a signal line 6, a motion compensation signal from a motion compensation prediction circuit 25 supplied via a signal line 81. After the size is reduced in accordance with the setting information signal of the size conversion rate, the signal is output to the binary image encoding circuit 22 via the signal line 82.
  • the resolution reduction conversion from the resolution conversion circuit 21 supplied via the signal line 2 a is performed.
  • the alpha-map signal (down-s amDled alDha-map signal) is encoded and output as a binary image.
  • the setting information signal of the size conversion rate supplied via the signal line 6 has a resolution of It is supplied to the conversion circuits 21, 23, 26 and the binary image coding circuit 22, and controls the amount of generated codes of the alpha-map signal. Is possible.
  • the code (setting information signal) of the size conversion rate supplied via the signal line 6 is coded by the multiplexing circuit 24 into an encoded alphanumeric circuit. The signal is multiplexed with the map signal, output via the signal line 3, and is encoded as an alpha-map encoded signal. It is provided to the multiplexing circuit 18 in FIG. 1 which is the final output stage of the image encoding device.
  • the resolution conversion circuit 21 is configured such that the alpha-map input via the alpha-map signal input line 2 is used.
  • the up signal is down-sampled in accordance with the setting information of a desired size conversion rate provided via the signal line 6, and the downsized analog signal is reduced. (Down-sampled alpha-map signal) is given to the two-direct image encoding circuit 22.
  • the binary image coding circuit 22 is a down-sampled alpha-map si ⁇ na 1, which is a resolution-reduced converted image signal obtained from the resolution conversion circuit 21. utilizing 3 ⁇ 4R N reduced resolution conversion circuits 2 6 or Shin Luo line 8 motion-out compensation prediction signal resolution reduction conversion a Le off ⁇ maps signal that will be supplied by through 2 for processing 3 ⁇ 4
  • the multiplexed encoded alpha-map signal which is the power, and the information of the magnification provided via signal line 6 are multiplexed. And output to signal line 3.
  • the resolution conversion circuit 23 converts the reduced-coded half-map signal (down-sampled / encoded alpha-map signal (binary image encoded output) is converted to the original resolution according to the size conversion rate setting information signal obtained via signal line 6.
  • the signal is converted to a full-map signal of the same degree, and the decoded signal is moved as a local decoded signal via the signal line 4 to the compensation prediction circuit 25.
  • the motion compensation prediction circuit 25 is provided with a frame memory, and the frame memory is provided with a resolution conversion circuit 23 for enlargement processing.
  • the supplied encoded video frame (revious encoded video frame signa 1) power S is stored.
  • the motion compensation prediction circuit 25 generates an alpha map motion compensation prediction signal in accordance with a separately supplied motion vector signal, and generates a signal line 8 1
  • the data is supplied to a resolution conversion circuit 26 for reduction processing via the.
  • the resolution conversion circuit 26 converts the supplied motion compensation prediction signal according to the size conversion rate setting information signal obtained via the signal line 6 to reduce the resolution.
  • the binary image encoding circuit 22 Provided to the binary image encoding circuit 22.
  • the binary image encoding circuit 22 is a motion compensation prediction signal of the resolution-converted alpha map provided from the resolution conversion circuit 26 for reduction processing. Is used to encode the resolution-reduced and converted alpha-map signal obtained from the resolution conversion circuit 21.
  • the decoding circuit 40 includes a binary image decoding circuit 41, a resolution conversion circuit (enlargement processing circuit) 42, a demultiplexing circuit 43, and a motion compensation prediction circuit. 44 and a resolution conversion circuit (reduction processing circuit) 45.
  • the demultiplexing circuit 43 is separated by the demultiplexing circuit 30 in the image decoding apparatus shown in FIG. 2, and is output to the alpha-map decoding circuit 40.
  • This circuit separates the input alpha-map signal into a code for the alpha-map signal and a code for the size conversion rate.
  • the conversion circuit 41 separates the code of the alpha-map signal from the separation circuit 43 and gives the code of the size conversion rate (the setting of the size conversion rate).
  • This is a circuit for returning to a binary image according to the constant information signal. The details will be described later, but it is supplied from the resolution conversion circuit 45 for reduction processing to the signal line 95.
  • the decoding is performed by using the motion compensation prediction signal of the resolution-converted alpha map.
  • the resolution conversion circuit 42 for enlargement processing converts the binary image, which is the code of the alpha-map signal from the binary image decoding circuit 41, into a separation circuit 4. According to the size conversion rate code (size conversion rate setting information signal) which is given separately from the three colors, the resolution is enlarged and output.
  • the configuration of the decryption circuit in the third embodiment is different from that of the first embodiment in that a motion compensation prediction circuit 44 and a resolution conversion circuit 45 for reduction processing are provided. This is different from the decryption circuit of 4.
  • the motion compensation prediction circuit 4 4 is based on the previously decrypted frame. It has a frame memory for accumulating playback images. It stores the playback signal supplied from the resolution conversion circuit 42 for enlargement processing, and also operates.
  • the vector signal (not shown) is supplied with the power S, generates a motion compensation prediction signal according to the motion vector signal, and reduces the signal via the signal line 94.
  • the configuration is such that it is supplied to a resolution conversion circuit 45 for processing.
  • the resolution conversion circuit 45 reduces the motion compensation prediction signal in accordance with the size conversion rate setting information signal supplied via the signal line 92, and then converts the signal.
  • the signal is output to the binary image decoding circuit 41 via the signal line 95.
  • the signal is supplied to the Alpha-map decoding circuit 40 via the signal line 8.
  • the separated codes are separated by a separation circuit 43 into a code of the file map signal and a code of the size conversion rate, and each of the signal lines 91 and The signal is output via the signal line 92.
  • the resolution reduction conversion supplied to the signal line 95 from the resolution conversion circuit 45 for reduction processing is performed.
  • the code of the alpha-map signal supplied via the signal line 91 and the supply via the signal line 92 are provided.
  • the data is reduced in size. Regenerates the Alpha-map signal and supplies it to the resolution conversion circuit 42 via the signal line 93 You
  • the reduced alphabet map signal reproduced from the binary image decoding circuit 41 is supplied via a signal line 92. Based on the sign of the conversion rate, the original size is expanded to reproduce the alpha map signal, and then output via the signal line 9.
  • the resolution conversion circuit 42 is composed of a reduced-encoded alpha-map signal (binary image data) provided from the binary image decoding circuit 41.
  • the image encoded output) is decoded to the original resolution in accordance with the setting information signal of the size conversion rate obtained via the signal line 92, and the local decoded signal is decoded. Then, the obtained local decoded signal is moved and output to the compensation prediction circuit 44.
  • a frame memory is provided in the motion compensation prediction circuit 44, and the frame memory is supplied from the resolution conversion circuit 42 for enlargement processing.
  • the motion compensation prediction circuit 44 generates a motion compensation prediction signal of the alpha map according to a separately supplied motion vector signal, and connects the signal line 94 to the signal line 94. Then, it is supplied to a resolution conversion circuit 45 for reduction processing.
  • the resolution conversion circuit 45 converts the supplied motion compensation prediction signal to a resolution reduction according to the size conversion rate setting information signal obtained via the signal line 92, Give to binary image decoding circuit 41.
  • the binary image decoding circuit 41 receives the motion compensated prediction signal of the resolution-converted alpha map given from the resolution conversion circuit 45 for the reduction process. Utilization and separation circuit According to the setting information signal of the size conversion rate from 4 3, the demultiplexing circuit 4 3 As described above, which is an outline of the decoding circuit to which the present invention is applied, the coding circuit in the third embodiment to which the present invention is applied. The configuration of this circuit is different from that of the encoding circuit in FIG. 3 in that it has a motion compensation prediction circuit 25 and a reduction circuit 26, and the decoding circuit is also different. Is different from the decryption circuit of FIG. 4 in that a motion compensation prediction circuit 44 and a reduction circuit 45 are provided.
  • the motion compensation prediction circuit 25 or 44 is provided with a frame memory for storing the reproduced image of the previously encoded frame.
  • the reproduced signal supplied from the expansion circuit 23 or 42 is stored.
  • the motion compensation prediction circuit 25 or 44 is supplied with a signal, a signal and a motion vector signal as shown in FIG.
  • a motion compensation prediction signal is generated in accordance with the vector signal and supplied to the reduction circuit 26 or 45 via the signal lines 81 and 94. Therefore, the motion vector signal is used in the motion compensation prediction circuit 11 or 35 provided in the apparatus shown in FIGS. 1 and 2. Even if the motion vector signal is used for IJ, the signal is good, and the motion signal for the alpha map is provided in the alpha map code circuit 20.
  • the ability to set up a torque detection circuit allows the motion vector signal for the alpha map to be determined.
  • the motion compensation signals supplied via the signal lines 81 and 94 are replaced by the signal lines 6 and 4 respectively.
  • the signal is output via the signal line 42 and the signal line 92.
  • the resolution-converted alpha-map signal supplied via the signal line 2 a is supplied. Is encoded as a binary image and output.
  • the point that the binary image encoding circuit 22 according to the present embodiment is fundamentally different from the encoding circuit of FIG. 3 is that the binary image encoding circuit 22 is supplied via a signal line 82. It has a function of encoding by using the motion compensated prediction signal of the resolution-converted alpha map.
  • FIGS. 16A and 16B are diagrams illustrating a coding method using a motion compensation prediction signal, and are illustrated in an image of a frame image unit. One of the divided NXM pixel configuration image blocks is shown.
  • block is the block to be processed, which is the block of the input current processed image.
  • the "compensated block” is a block that compensates for the image that was previously processed.
  • the reference change pixel bl on the block of the alphabet corresponding to the block of the current processing target image is changed to the change pixel a0, It was detected in the same "current block” as a1.
  • the reference change pixel b 1 is a motion compensation prediction signal.
  • the reference change pixel b1 on the block of the alphabet map corresponding to the block of the current processing target image is determined by the motion compensation prediction signal. Detect and detect from within a compensated block.
  • This embodiment is different from the first embodiment only in the method of detecting the reference change pixel b1, except that the encoding is performed by using the relative addresses of a0, a1 and b1.
  • ⁇ Decryption is the same as in the previous embodiment.
  • a0 is the origin changing pixel, and the encoding is already completed up to the origin changing pixel a0.
  • a1 is the next transformed pixel after the starting point changed pixel a0, and bO is in the "compensated block" and has the same position as a0 (the changed pixel and the changed pixel). Is limited,).
  • the reference change pixel b1 is defined as follows.
  • abs—X is the address of pixel X when the block is scanned in the raster order from the upper left pixel card in the block. The address of the pixel at the top left of the block is
  • Figure 16A shows the case where the changed pixel is not on "a0-1ine". In this case, the first changed pixel in the next line is
  • FIG. 16B shows a case where the changed pixel is on “a0-1ine”, but since this changed pixel X is not the opposite color to “a”, “b” Instead of "1", set the first changed pixel of the next line as "b1".
  • the encoding method described with reference to FIGS. 16A and 16B is an example of a method for obtaining a reference compensated pixel b 1 from a “compensated block” internal force.
  • various deformations are possible.
  • the binary image decoding circuit 41 is provided with a motion compensation prediction signal (“compens”) of the resolution-reduced and converted alpha map supplied through the signal line 95. ated block "), the reference change pixel b1 can be detected in the same procedure as the binary image coding circuit 22.
  • a motion compensation prediction signal (“compens”) of the resolution-reduced and converted alpha map supplied through the signal line 95. ated block ")
  • the reference change pixel b1 can be detected in the same procedure as the binary image coding circuit 22.
  • the power for detecting the reference change pixel b1 from the "current block” and the power for detecting the reference change pixel b1 from the "compensated block” are cut off, for example, in units of blocks. It can be replaced.
  • information for switching is also encoded in the binary image encoding circuit 22, and information for switching is also encoded in the binary image decoding circuit 41.
  • the reference change pixel 1 is detected from the "current block” on the basis of the switching information, and the power is detected from the "compensated block".
  • the detection force should be switched, for example, on a block-by-block basis.
  • a means for switching the scan order may be provided to switch the scan order to a horizontal scan as shown in FIG. 17A.
  • a vertical scan as shown in Fig. 17B, the number of changing pixels is reduced, and the code amount is further reduced. Thus, this also leads to more efficient encoding.
  • a moving image code for encoding a plurality of frames of moving image signals obtained as time-series data for each object having an arbitrary shape is described.
  • an encoding circuit for encoding an alphabet map representing the shape of an object in an encoding device.
  • the rectangular area including the object is divided into MXN pixels (M: the number of pixels in the horizontal direction of the water, N: the number of pixels in the vertical direction).
  • the resulting blocks can be sequentially and coded according to a regular rule within a rectangular area, and all or some of the blocks can be encoded.
  • An encoder that performs relative address encoding on the other hand, and a memory that stores the playback value near the block, and a frame that has already been encoded A frame memory for storing a reproduction signal of the frame, a motion compensation prediction circuit for generating a motion compensation prediction value using the reproduction signal in the frame memory, and It has a detection circuit that detects changed pixels including the reproduction value near the block, and has a relative address encoding. Change the reference change pixel to a pixel in the block
  • a moving picture coding apparatus is provided which is obtained from a motion compensation prediction signal instead of a value.
  • the alpha-map decoding circuit for each block composed of MXN pixels, the inside of the rectangular area including the object is regularly arranged.
  • a decoding circuit that decodes sequentially, a memory that stores the playback value near the block, and a playback signal of the frame that has already been encoded.
  • a motion compensation prediction circuit that generates a motion compensation prediction value using the frame memory, a reproduction signal in the frame memory, and a reproduction value in the vicinity of the block.
  • a detection circuit for detecting a change pixel including the reference change pixel in the relative address encoding, not from a pixel value in the block.
  • a video encoding device that is obtained from a motion compensation prediction signal is provided.
  • a playback value storage circuit for storing a playback value in the vicinity of the block, and a frame memory for storing a playback signal of an already encoded frame.
  • a detection circuit that detects changed pixels including the reproduction value near the block, and a relative address code obtained from the reproduction pixel value in the block Reference change pixel and motion compensation prediction signal
  • a switching circuit for switching the reference change pixel obtained from the reference address encoding.
  • the relative address encoding information is output by the switching circuit.
  • a moving picture coding device for coding together with the replacement information is provided.
  • a fixed area within a rectangular area including an object is fixed.
  • a playback value storage circuit that stores the playback value near the block, and stores the playback signal of the frame that has already been encoded.
  • a motion compensation prediction circuit that generates a motion compensation prediction value using the reproduced signal in the frame memory and a reproduction signal in the frame memory. It has a detection circuit that detects changed pixels including the playback value, and has a relative address coding obtained from the playback pixels in the block. And the reference change pixel of the relative address encoding determined from the motion compensation prediction signal.
  • the reference change pixel b1 is detected from within the "current block", and is detected from within the compensal: ed block.
  • the power can be switched and processed in block units, and the encoding side also encodes the information for this switching, and the decoding side. In this case, this is decoded and based on the information for switching at the time of decoding processing of the encoded alphabet.
  • the reference change pixel b1 is detected from the "current block” or the power to be detected from the "compensated block” is switched in block units. In this way, the optimal processing can be performed based on the image content in block units, and the efficiency can be further improved. Encoding becomes possible.
  • the information of the alphabet map which is the sub-image information indicating the shape of the object and the position in the screen, is efficiently encoded.
  • an image encoding apparatus and an image decoding apparatus capable of performing the decoding can be obtained.
  • the alpha-map encoding circuit 20 uses an MMR (Modified Modified READ) encoding.
  • MMR Modified Modified READ
  • the present invention is not limited to MMR coding, and can be realized using any other binary image coding circuit. Such an example is described below.
  • the alpha-map coding circuit 20 and the alpha-map decoding circuit 40 are described. Explain the specific configuration.
  • Fig. 18 shows an example of the screen of an alpha-map, which is divided into macro-block (MB) units of a predetermined multiple pixel configuration such as 16 XI 6 pixels.
  • a square cell is a dividing line, and each cell is a macro block (MB).
  • MB macro block
  • the shape information of the object is represented by one of the values representing opaque, which is transparent for each pixel. Therefore, as shown in FIG. 18, the state of the contents of each macro block (MB) on the alpha-map screen is “transparent”.
  • the macroblock (MB) will have the following four code modes.
  • the alpha-map encoding circuit 20 and alpha-map decoding are used for the encoding or decoding of each mode. It will be explained in the coding circuit 40.
  • FIG. 19 is a detailed configuration diagram of the alphabet encoding circuit 20.
  • the mode determination circuit 110 the mode determination circuit 110, CR (size conversion rate) determination circuit 111, selector 120, block Pixel value setting circuit 140,
  • the pixel value setting circuit 140 in the block generates pixel data that makes all the pixels in the macroblock transparent.
  • the pixel value setting circuit 150 in the block is a circuit that generates pixel data that makes all pixel values in the macro block opaque. It is.
  • the CR (size conversion rate) determination circuit 1 1 1 analyzes an Alpha-map signal supplied through the Alpha-map signal input line 2 and Alpha-map images for frames Is determined at what size conversion ratio should be processed, and the result is output as the size conversion ratio (conversion ratio) b2. Further, the reduction circuit 17 1 receives the file map signal supplied via the alpha map signal input line 2 for the entire frame.
  • the scan type (ST) determination circuit 177 determines the scan direction based on the encoded output from the binary image encoding circuit 170. And output scan direction information (scan type) b4.
  • the transposition circuit 175 is based on the scan direction information b4 output from the scan type (ST) determination circuit 177, and the reduction circuit 1 7
  • the position of each macroblock is transposed with respect to the one-frame analog map signal which has been reduced by 1 and the transposition circuit 1 7 6
  • the outputs of the reduction circuits 171, 172, and 173 are transposed based on the scan direction information b4.
  • the binary image encoding circuit 170 encodes the reduced analog map signal supplied via these transposition circuits 1775 and 1776. Output.
  • the expansion circuit 172 outputs the alpha-map signal given via the reduction circuit 171 to the CR judgment circuit 11 1.
  • the motion compensation prediction circuit 160 performs the enlargement processing at the noise conversion rate, and operates using the reproduction image of the reference frame stored in the frame memory 130.
  • a compensation prediction signal is generated and output to the mode determination circuit 110 and the reduction circuit 174.
  • the reduction circuit 1 7 4 uses the motion compensation prediction signal.
  • the reduction processing is performed at the size conversion rate output from the CR judgment circuit 1 1 1, and the reduction circuit 1 73 is used to store the reference file stored in the frame memory 130.
  • the playback image of the frame is reduced at the size conversion rate output by the CR judgment circuit.
  • the selector 120 reproduces from the block pixel value setting circuit 140 in accordance with the classification information output from the mode determination circuit 110.
  • the frame memory 130 stores the output of the selector 120 in frame units.
  • the mode determination circuit 110 is a motion compensation prediction circuit.
  • the alpha map signal is obtained. Analyzes the alpha-map signal supplied via input line 2 and, for each macroblock,
  • Judgment is made on any of the categories classified as transparent, oaque, no update, and "coded", and the judgment result is output as mode information b0.
  • the motion vector detection circuit (MVE) 178 is driven by the alpha-map signal supplied via the alpha-map signal input line 2. Detect the vector.
  • the MV encoding circuit 179 encodes the motion vector detected by the motion vector detection circuit (MVE) 178, and encodes the encoded motion vector. Conclusion The result is output as motion vector information b1. For example, when prediction coding is applied to the MV coding circuit 179, the prediction error signal is output as the motion vector information b1. become .
  • VLC (Variable Length Coding) ⁇ Multiplexing circuit 180 is the mode information b 0 from the mode determination circuit 110 and the MV coding circuit 179
  • the motion vector information b 1 and the size conversion rate information from the CR (size conversion rate) judgment circuit 11 1 b 2 and the scanner Type (ST) decision circuit 177 Receives scan direction information b3 from input device and binary image coding circuit b4 and receives binary coded information b4 from binary image coding circuit 170. These are then subjected to variable-length coding, multiplexed, and output to the signal line 3.
  • the alpha-map signal to be encoded is transmitted via the alpha-map signal input line 2 to the alpha-map signal. Is supplied to the encoding circuit 20. Upon receiving this, the file map encoding circuit 20 analyzes the alpha map signal through the mode determination circuit 110. And the macro block 'transparent'
  • the mode determination circuit 110 is a miss when all the signals in the input macro block are transparently replaced. Calculate the number of matching pixels and classify macroblocks whose number is less than or equal to a threshold value as "transparent". Similarly, when the entire macro block is opaquely replaced, the number of mis-matched pixels becomes less than the threshold value. o aque.
  • the mode determination circuit 110 is
  • o aque for t5 remacroblocks, the mis-match between the motion compensation predicted value supplied via the signal line 101 and Calculates the number of pixels, and removes blocks that are less than the threshold value S
  • the classification information b0 obtained by the mode determination circuit 110 is supplied to the selector 120 via the signal line 102, and the classification information b0 is supplied to the selector 120. Is the mode of the block
  • the reproduction signal m O in which all the pixel values in the block are transparent in the block pixel direct setting circuit 140 is selected, and the signal is transmitted. It is supplied to the frame memory 130 via the line 4 and is stored in the storage area of the frame, and is also stored in the alpha-map encoding circuit 20. Output as the output of.
  • the selector 120 sets the pixel value in the macro block.
  • the setting circuit 150 select the reproduction signal m1 in which all the pixels in the macroblock are opaque, and select the macroblock.
  • the motion compensation prediction signal m generated by the motion compensation prediction circuit 160 supplied via the signal line 101 is used. If 2 is selected and the mode of the block is "coded”, the reproduction signal supplied via the reduction circuit 1 11 and the expansion circuit 172 is supplied. No. m 3 is selected, supplied to the frame memory 130 via the signal line 4, stored in the storage area of the frame, and stored in the memory. Output as the output signal of the famap encoding circuit 20.
  • the pixel value of a macro block classified as “coded” in the mode determination circuit 110 is reduced in the reduction circuit 171. After that, it is encoded in a binary image encoding circuit 170.
  • the size conversion rate used in the reduction circuit 17 1 (Conversion
  • Ratio (size conversion rate) setting information is obtained by the CR judgment circuit 111. For example, if the reduction ratio is "1 (do not reduce), ...,” 1/2 (1Z2 for both horizontal and vertical), “1/4 (1/4 for both horizontal and vertical) ", The CR determination circuit 111 determines the CR (size conversion rate) in the next step.
  • the reduction rate of the macro block is set to “1,”.
  • the value of the CR (size conversion rate) obtained in this way is reduced via the signal line 103 to the reduction circuits 17 1, 17 3 17 4, and the enlargement circuit. 17 2 and supplied to a binary image coding circuit 170 and also supplied to a VLC (variable length coding) and multiplexing circuit 180 to be coded. Later, it is multiplexed with other codes. Further, in the transposition circuit 1775, the position of the signal of the reduced block supplied from the reduction circuit 1771 is transposed (addition in the direction of water square). The dress and the vertical address are swapped.)
  • the encoding is performed in the order of the scan in the horizontal direction of water (horizonta L scan), and in the order of the scan in the vertical direction (vertica scan). ⁇ ; ⁇ is switched.
  • the transposition circuit 1776 the reproduction pixel value in the vicinity of the macro block reduced by the reduction circuit 1773 and the reduction pixel 1774 are used. With the reduced motion compensation prediction signal, the position of each signal is transposed.
  • the determination of whether or not to perform the transposition processing is performed, for example, by using a horizontal image coding circuit 170 in horizontal transposition.
  • the scan and the vertical scan are respectively encoded, and the encoded information b4 output via the signal line 104 is determined as ST (Scan Type).
  • the ST determination circuit 177 can select the power S and the scan direction in which the code amount is reduced. Determined by
  • the macroblock signal supplied from the transposition circuit 175 is supplied from the transposition circuit 176. Encode using the reference signal.
  • the method used in the above-described third embodiment can be cited.
  • the present invention is not limited to this.
  • another binary image encoding can be applied to the binary image encoding circuit 170.
  • intra coding mode using the reference pixel in the frame
  • inter referring to the motion compensation prediction signal.
  • intra / inter is performed by, for example, a binary image coding circuit supplied via a line 104.
  • the encoding information from 170 can be supplied to the mode determination circuit 110 to select the mode that reduces the amount of code.
  • information on the selected encoding mode (intra / inter) is supplied to the binary image encoding circuit 170 via the signal line 105. .
  • the coding information of the binary image coding circuit 170 which has been coded in the optimal mode selected by the above means, is transmitted through the signal line 104.
  • the signal is supplied to the VLC 'multiplexing circuit 180 via the VLC' and is multiplexed with other codes.
  • information on the optimum mode is supplied from the mode determination circuit 110 to the VLC / multiplexing circuit 180 via the signal line 106 and is encoded. After that, it is multiplexed with other codes.
  • the motion vector detection circuit (MVE) 178 the optimum motion vector is detected.
  • the detected motion vector is supplied to a motion compensation prediction circuit 160 via a signal line 107, and is also supplied to an MV encoding circuit 179. After being coded, it is supplied to a VLC / multiplexing circuit 180, and after being coded, multiplexed with another code.
  • the multiplexed code is output via signal line 3.
  • the motion compensation signal is stored in the frame memory 130 based on the motion vector signal supplied via the signal line 107.
  • the motion compensation prediction signal is generated using the reproduced image of the reference frame, and the mode determination circuit 110 and the reduction circuit are generated via the signal line 101. Output to 1 7 4.
  • FIG. 20 is a detailed configuration diagram of the alpha-map decoding circuit 40.
  • the alpha-map decoding circuit 40 has a VLC (variable-length coding) / separation circuit 210, a mode reproduction circuit 220, and a cell-selection circuit. 230, Pixel value setting circuit in macro block 24, 250, Motion compensation prediction circuit 260, Frame memory 27, Binary image Decoding circuit 280, Enlargement circuit 281, Transposition circuit 282, 285, Reduction circuit 283, 284, Binary image decoding circuit 280 And a motion vector reproduction circuit 290.
  • VLC variable-length coding
  • VLC Very Length Coding
  • demultiplexing circuit 210 are used to encode the encoded bits of the multiplexed and transmitted Alphamap.
  • the stream is decoded, and the mode information b0, the motion vector information b1, the size conversion rate information b2, the scan direction information b3, and the binary image code
  • the mode reproduction circuit 220 receives the separated mode information b0, and outputs “transparent”, oaque, and no update. This is a circuit that reproduces any of the four coded modes of coded and coded.
  • the binary image decoding circuit 280 outputs the binary image coding information b 4 separated by the VLC (variable length coding) / separation circuit 210, Size conversion rate information b 2 and the mode played back from the mode playback circuit 220 — It is a circuit that decodes and outputs a binary image using the information of the code information and the information of the transposition circuit 285, and the enlarged circuit 281 is separated as described above. This is a circuit for enlarging the decoded binary image using the information of the size conversion rate information b2, and a transposition circuit 282 converts the image after the enlarging processing. This is a circuit that performs a transposition process according to the separated scan direction information b3 and outputs it as a reproduced signal m3.
  • the macroblock / pixel value setting circuit 240 generates a reproduced signal mO in which all pixel values in the macroblock are transparent.
  • the circuit is a circuit, and the pixel value setting circuit 250 in the macro block generates the reproduction signal m 1 in which all the pixel values in the macro block are opaque. It is a circuit that grows.
  • the motion vector reproduction circuit 290 uses the motion vector information b 1 separated by the VLC (variable length coding) / separation circuit 210. This is a circuit that reproduces the motion vector of the macro block, and the motion compensation prediction circuit 260 uses this reproduced motion vector.
  • the stored reference frame of the frame memory 270 is moved from the reproduced image to generate a compensated predicted value m 2, and the reduction circuit 284 is a circuit for generating the compensated predicted value m 2.
  • the motion compensation predicted value m 2 is reduced using the information of the obtained size conversion rate information b 2, and the reduced circuit 28 3 is used for the separated size.
  • the image of the reference frame stored in the frame memory 270 is reduced using the information of the conversion rate information b 2, and the transposition is performed.
  • Circuit 285 is the VLC (Variable Length Coding) Based on the scan direction information b3 separated by 210, the reduced images output from the reduction circuits 283 and 284 are transposed and binary image decoding is performed. This is a circuit that outputs to the synthesis circuit 280.
  • the selector 230 sets the pixel value setting circuit 2 in the mouth block according to the mode information output from the mode reproduction circuit 220. Either of the reproduced signals mO and ml of 40 and 250, the predicted value of motion compensation m2 of the motion compensation prediction circuit 260, and the reproduction signal m3 of the transposition circuit 2882.
  • the frame memory 270 outputs the image signal output from the selector 233 on a frame basis. This is the memory to keep.
  • the alpha-map decoding circuit 40 having such a configuration is supplied with the alpha-map encoding bitstream via the signal line 8. It is done.
  • This bitstream is supplied to a VLD (Variable Length Decoding) / separation circuit 210. Then, in the VLD (Variable Length Decoding) / separation circuit 210, this bit stream is decoded, and the power, mode information b0, and motion
  • the vector information is separated into vector information b1, size conversion rate information b2, scan direction information b3, and binary image coding information b4.
  • the separated information is managed on a macroblock basis.
  • the mode information b0 is supplied to the mode reproduction loop 220, and the macroblock is set to any of the following modes. Classify whether it belongs to. (1) transparent "
  • the "coded” includes the mode of "intra” and the mode of "inter” as described in the embodiment of the encoding circuit in FIG.
  • the reference pixel in the frame is used in the "intra” encoding mode, and the reference to the motion compensation prediction signal is used. inter "encoding mode.
  • the mode of the macro-block is changed to “transparent”.
  • the reproduction signal m0 in which all the pixel values in the mask block are made transparent in the block pixel value setting circuit 240 is selected, and the signal is transmitted.
  • the frame memory 270 is supplied via the line 9 to the frame memory 270, and the image belonging to the macro blocker S of the frame memory 270 belongs to the frame memory 270.
  • An image of the alpha-map that is stored in the frame storage area and decoded from the alpha-map decoding circuit 40 is also stored. Output.
  • the pixel value setting circuit 250 in the macro block sets the mode in the block.
  • the reproduction signal m1 whose pixel values are all opaque is selected, and when the mode of the block is "no update", the reproduction signal m1 is supplied via the signal line 202.
  • the motion compensation prediction signal m 2 generated in the supplied motion compensation prediction circuit 260 is selected, and the mode of the block is “ In the case of "coded”, the reproduction signal m3 supplied through the enlargement circuit 281 and the transposition circuit 282 is selected, and the frame memory is transmitted via the signal line 9. 270, and is stored in a storage area of an image frame to which the macro block of the frame memory 270 belongs. It is output as an image of the decoded alpha-map from the alpha-map decoding circuit 40.
  • the separated motion vector information b 1 is supplied to the motion vector reproduction circuit 29, and the motion vector of the macro block is Will be played.
  • the regenerated motion vector is supplied to the motion compensation prediction circuit 260 through the signal line 203, and the reproduced motion vector is supplied to the motion compensation prediction circuit 260.
  • the motion compensation prediction value is obtained from the reproduction plane image of the reference frame stored in the frame memory 270. Generate.
  • the signal is output to the reduction circuit 2884 and the selector 2330 via the signal line 202.
  • the separated size conversion rate information b 2 is supplied to the reduction circuit 283 and the enlargement circuit 281, and the binary image decoding circuit 2 80, and the scan direction information b3 is supplied to the transposition circuits 2822 and 2885.
  • the transposition circuit 282 that has received the scan direction information b3, the macroblock returned to the original size, which is supplied from the expansion circuit 281, is provided.
  • the position of the playback signal is transposed (water The square address and the vertical address are swapped).
  • the transposition circuit 285 which has received the scan direction information b 3, the reproduction pixel value near the block reduced by the reduction circuit 283 and the reduction pixel value The position of each signal is transposed with the motion compensation prediction signal reduced by the circuit 284.
  • the two macroblocks separated by the separating circuit 210 are The value image encoding information b4 is decoded using the reference signal supplied from the transposition circuit 2885.
  • the output of the enlargement circuits 172 and 281 has image quality deterioration due to discontinuity in the oblique direction.
  • the expansion circuits 17 2 and 28 1 are equipped with filters to suppress the discontinuity in the oblique direction. You can get it.
  • the configuration of the binary image decoding circuit 280 is the same as the configuration example shown in the third embodiment described above. Not exclusively.
  • the embodiments of the binary image encoding circuit 170 and the binary image decoding circuit 280 are not limited to the above-described third embodiment. , Was described. Here are some other examples.
  • Figure 21 illustrates an example of the Markov model coding method.
  • the pixel X in the figure is the target pixel to be encoded, and the pixels a to f are reference pixels that are referred to when encoding the pixel X.
  • the pixels a to f are pixels that have already been encoded when the pixel X is encoded.
  • the variable pixel code table is composed of an arithmetic code (arithmetic code).
  • the probabilistic taper (probability table) power S is adaptively switched and encoded.
  • the "intra" encoding is performed before the destination pixel in the macroblock that is currently being processed.
  • the encoded pixel is taken as the reference pixel, and the "inter” encoding is coded before the target pixel in the macroblock currently being processed. It is only necessary that the pixels in the motion compensation prediction error signal be used as reference pixels in addition to the pixels that have been set.
  • [method 1] the method of the third embodiment (encoding method based on MMR) and [method 2] Markov model coding are applied. It is characterized by having both types of switching and encoding methods, and adopting a form in which these are adaptively selected and used.
  • FIG. 22A shows a binary image code used in this embodiment. It is a block diagram of the chemical circuit 170. As shown in the figure, the binary image encoding circuit 170 is composed of a pair of selectors 421 and 424 and a pair of binary image encoding circuits. ing . Of these, the selector 421 is a selection circuit on the input side, and the selector 422 is a selection circuit on the output side.
  • the binary image encoding circuit 42 2 is a first binary image encoding circuit to be encoded by the above [method 1], and the binary image encoding circuit 42 3 is This is the second binary image encoding circuit encoded by the method 2].
  • the signal of the processing target macro block supplied from the transposition circuit 175 via the signal line 411 is transmitted to the signal processor 407.
  • the first binary image code is selected by the selectors 42 1 and 42 4 that are selectively switched according to the switching signal supplied via the signal line 4 13.
  • the second binarized image coding circuit 4 2 3 is adaptively distributed, and the separated macroblock signal is encoded. Encoding.
  • the encoded information is output via signal line 412.
  • the signal lines 103 and 105 of the signal supplied to the binary image encoding circuit 170 and the signal lines of the transposition circuit 176 are , Has been omitted.
  • the switching information supplied via the signal line 4 13 may be a preset default value or may be appropriately determined based on the image content. In order to obtain the necessary switching information, encode it separately as side information, and transmit it to the decoding side. You can do it.
  • FIG. 22A is a block diagram of a binary image decoding circuit 280 used in the present embodiment.
  • the binary image decoding circuit 280 is composed of a pair of selectors 441 and 444, and a pair of binary image decoding circuits 442 and 444. It is composed of 3 and power.
  • the selector is composed of 3 and power.
  • Reference numeral 441 denotes a selection circuit on the input side
  • selector 444 denotes a selection circuit on the output side.
  • the binary image decoding circuit 44 is a first binary image decoding circuit for decoding the image encoded by the above [method 1], and is a binary image decoding circuit.
  • the decoding circuit 443 is a second binary image decoding circuit that decodes the [method 2] encoded image.
  • the binary image decoding circuit 280 having such a configuration, the binary image code of the macro block to be processed, which is rendered via the signal line 431, is provided.
  • the conversion information b 4 is converted into the first binary image by the selectors 4 4 1 and 4 4 4 according to the switching signal supplied via the signal line 4 3 3.
  • the signal is adaptively distributed to the decoding circuit 442 and the second binary image decoding circuit 443 for decoding.
  • the decoded signal is output via a signal line 432.
  • the binary image decoding circuit 280 The signal line for supplying the signal from the demultiplexing circuit 210 to the mode regenerating circuit 220, the signal line from the transposing circuit 285, and the signal lines for these are omitted. It is.
  • the switching information supplied via the signal line 433 may be a default value, or may be a coded transmitted bit.
  • the information sent from the encoder that obtains the switching information from the stream may be used.
  • the present invention is based on the video obect plane
  • V0P Video Processing Unit
  • Pex is a pixel position after conversion, and Pex indicates a real pixel position as shown in FIG. 23A.
  • Such a process is a process called “bilinear interpolation”, and the pixel direct Ia to Id power of A to D, and the pixel direct I ⁇ of Pex can be easily obtained. Can be obtained.
  • FIG. 24A is a binary image of the original size
  • FIG. 24B is a binary image obtained by reducing the image.
  • the object area is indicated by a black circle
  • the knock-ground (background) area is indicated by a white circle.
  • the sampling transformation (enlargement / reduction transformation) is performed to smooth out the discontinuity in the oblique direction that occurs.
  • the pixels in the knock ground area are smooth out.
  • the pixel at the position indicated by the double circle (that is, the circle)
  • the pixel to be inspected) is set as a pixel in the black circle, and is set as a pixel in the object area.
  • the pixel with a black circle is "1”. If the white circle is "0", the pixel at the position indicated by the double circle (pixel value "0") has a pixel value of "1". Replace with ".
  • FIG. 25 shows another example of a smoothing filter (smoothing filter).
  • the center pixel of the 3 ⁇ 3 pixel mask shown in FIG. 25 is C, the upper left pixel is TL, the upper right pixel is TR, and the lower left pixel is C with respect to C.
  • the filtered value of pixel C is determined by the following equation.
  • the value of the object is represented by "1" and the value of the background is represented by "0".
  • the value obtained by caloring the pixels TL, TR, BL, and BR is greater than “2”.
  • the value of pixel C is set to "1” if it is large, and pixel C is set to "0" if it is not large. If S is not "0”, check whether the value obtained by force-calculating the pixels TL, TR, BL, and BR is smaller than "2". If smaller, check the value of pixel C.
  • the configuration of the smoothing filter is not limited to the above example.
  • nonlinear filter such as f ilter
  • Figure 26A shows that both horizontal and vertical are doubled by bilinear interpolation. It is a figure showing the process which expands to.
  • FIG. 26B the pixels of the reduced block (decoded pixels) are represented by white circles, that is, "o" marks, and the pixels to be interpolated.
  • I p 1 I a
  • I ⁇ 2 I b
  • I D 3 I c
  • I p 4 I d
  • the enlarged image has the same value of 2 ⁇ 2 pixels, and the smoothness is low. Lost .
  • the above problem is solved by changing the weighting of the bilinear interpolation as follows.
  • Pixel value (“1" or "0").
  • A, B, C, and D are the pixels of the reduced block, and Ia, Ib, Ic, and Id are the pixels of the pixels A, B, C, and D.
  • the pixel value I p2 of the pixel p 2 is set to “1” if the sum of the double value of lb, la, I c, and I d is larger than 2, and so on. Otherwise, it indicates that "0" is assumed.
  • the pixel Ip3 of the pixel ⁇ 3 is defined as "1" if it is greater than the sum S2 of twice the Ic, la, lb, and Id. , Otherwise it is assumed to be "0",
  • the pixel value I p4 of the pixel p 4 is set to “1” if the sum of twice the value of I d, I a, I b, and I c is 2 or more, and If it doesn't, "0" will be shown.
  • a table is prepared and stored in the memory, and pixels are uniquely replaced according to the table. It is.
  • the address of the memory is 4 bits and the addresses obtained by arranging Ia, Ib, Ic, and Id are given as Ia , I b, I c, I d.
  • the Ipi, Ip2, Ip3, and Ip4 obtained by the turn should be recorded in advance as shown in the table below.
  • I p 2 I p 3, I p 4, and the power “0, 0, 0, 0” are defined as “I a 1 b, I c, I d, force” 0, 0, 1 ,
  • the contents of such a table are represented by la, Ib, Ic, and Id are addresses, and the data stored in the addresses are Ipl, Ip2, Ip3, and Ip4.
  • the addresses of la Ib, Ic, and Id are input into the memory so that the settings are retained in the memory so that Then, by simply reading out the corresponding Ip1, Ip2, Ip3, and Ip4, it is possible to obtain the internal value S.
  • the sequence of Ia, Ib, Ic, and Id is a binary number, and the number converted to a decimal number is called a context.
  • the internal values I pi, I p2, I p3, and I d are obtained from the contexts obtained by I a, I b, I c, and I d.
  • the number of pixels to be referred to was four.
  • the method is not limited to this and is explained below. It can be realized in any number of cases, such as 12 or more.
  • the context (20 to 51 1) is obtained as the reference pixel for the 9 pixels of the image area, and the inner pixel P 1 is determined according to the context. There is also a method to determine whether it is "0" or "1".
  • the pixels P2, P3, and P4 the interpolation position of each pixel and the immediate surroundings of that position According to the positional relationship of the four reference pixels, the pixels surrounded by the dotted lines in FIGS. 27B, 27C and 27D are used as the reference pixels.
  • the reference pixels A to I are arranged so that the relative positions with respect to the inner pixel P are the same (for example, FIG. 27 7 2 can be rotated 90 degrees clockwise so that the same pixel can be obtained. Rotating the turn only results in the same index, so the memory used as a table with common contents for the interpolated pixels P1 to P4 is used. You can use it.
  • External reference pixel A T is 2 rows and 8 columns, external reference pixel
  • border A L is an array of 4 rows and 2 columns
  • an external reference pixel BT is an array of 2 rows and 12 columns
  • an external reference pixel B L is an array of 8 rows and 2 columns.
  • these external reference pixels are calculated by averaging the pixels at a given position in the macroblock that has already been reproduced. Must be obtained as a value.
  • an external Referencing the reference pixels AT and AL when encoding with the size of 8 x 8 pixels, if referencing the external reference pixels BT and BL, use these to enlarge. It is possible to avoid the waste of requiring an external reference pixel only for processing.
  • the average value is not obtained, and the external reference pixels AT and AL are used to determine the language for the combi- nation programming. If written in C language
  • Figure 30 shows an example of a reduction process that reduces a block (macro block) to the size of "1Z2, '" in the vertical and horizontal directions.
  • the area within the dotted frame is shown. If each area is defined as a unit reduction block area, 2 ⁇ 2 pixels within the unit reduction block area (total indicated by “O” in each dotted frame) The average value ("X" in Fig. 30) for each 4 pixels) is taken as the pixel value in the unit reduction block area, that is, the macro block is used. If you want to reduce the size to "14" vertically and horizontally, For each unit reduction block area, the average value for each 4 ⁇ 4 pixel in the area is obtained, and the average value is obtained as the pixel value in the unit reduction block area. .
  • the value of pixel X in the unit reduced block area is considered.
  • the average value of pixels A to D is not calculated, but the pixel position in a wider range than this , K, L, M, N, O, and P, and the average of these A to P can be obtained. Good.
  • a method is adopted in which, among the pixels in the adjacent unit reduced block area, pixels adjacent to the pixels A to D are included, and these average values are adopted.
  • the average value of the existing pixel values is set as the pixel value of the unit reduction block area, so that the number of pixels of the unit reduction block area is reduced. In this case, the size of the macro block was reduced. Without such calculations, it is possible to reduce the size of a macro block even by mechanical thinning processing.
  • FIG. 32 is an example of a reduction process by pixel thinning.
  • the macro block MB with a 16 x 16 pixel configuration has been reduced to a block with an 8 x 8 pixel configuration as a result of reduction processing by pixel thinning. are doing . That is, the figure Pixels indicated by dotted white circles in the graph are pixels that have been thinned out, and pixels indicated by solid white circles represent the pixel values of the unit reduction block area (see FIG. 32). "X"). In this case, the reduction ratio (CR) indicates the ratio of pixels to be decimated.
  • the method of pixel thinning is not limited to the method described with reference to FIG. 32.
  • the grid of a one-of-dice CO ⁇ Pixels are thinned out in a) shape. In this case, the enlargement processing is equivalent to interpolating the thinned pixels.
  • FIG. 33 is a diagram illustrating the enlarging process.
  • the solid rectangular frame indicates a macro block
  • each cell in the portion indicated by a dotted frame indicates a unit reduction block area. ing .
  • the white circles in the unit reduction block area represent existing pixels (coded pixels), and the number of pixels is increased by interpolation to expand the macro block. Try to figure it out.
  • the interpolated pixel is
  • the pixel value outside the macro block is required.
  • the nearest pixel value in the macro block can be assigned.
  • the pixel interpolation of a certain unit reduction block area is performed.
  • the pixel value power s of a total of 9 pixels that is, the pixel values of the surrounding 8 blocks, which are the unit reduction block ⁇ -block regions of the self and the self itself, is necessary for the processing.
  • the unit reduced block area in the macro block is located at the boundary between the macro block and the macro block, the surrounding area is changed to 8 blocks. Some of them are blocks belonging to the ⁇ -block belonging to the ⁇ -block which does not belong to the user. There is a necessary power S to get.
  • the pixel value of the pixel closest to the pixel in the macroblock to which the user belongs belongs to each pixel. Allotment to the side blocks may be performed by using those pixels directly as the pixel values in the neighboring unit reduction block area necessary for processing the pixel values for convenience.
  • the size conversion process for each macro block does not need to be closed within the macro block, as shown in FIG. 29.
  • the reproduction value near the block (the pixel value in the block adjacent to the upper right and upper left corners and the upper right corner) may be used.
  • the solid rectangular frame indicates a certain block (mark ⁇ block), and the “X” mark indicates an image at 1 ⁇ magnification (standard magnification image). ) Is shown for each pixel.
  • a macro block usually has a 16 x 16 pixel configuration, with the frame compressed to 1 to 2 and the pixel in the block at the mouth is an 8 x 8 pixel configuration. And the previous 16 x 16 pixel configuration is respectively 9
  • the image of 2 X 2 pixels in this case is represented by representing the information of the representative point in a one-pixel representation format. This is the “ ⁇ ” mark in Fig. 29.
  • the frame shown by the dotted line is the unit reduced aperture area, which is a 4-pixel configuration (2X2) in the standard magnification image. ), And in the case of 1/2 reduction, each pixel within this dotted frame is represented by one pixel.
  • pixels 1 and 2 in a certain unit block block area (area within the dotted line frame) of a certain macro block are represented by inner lines.
  • pixels 1 and 2 in a certain unit block block area (area within the dotted line frame) of a certain macro block are represented by inner lines.
  • pixels 3 and 6 are pixels belonging to the adjacent block (adjacent macro block), and the pixel before enlargement is still
  • the average value of pixels 7, 8, 9, 9 and 10 which are existing neighboring pixels after interpolation in the macro block is obtained, and One method is to use the value of “pixel 3” that has been discarded. However, if the calculation for obtaining the average value is reduced, the pixel 7 is used. , 8, 9 and 10 out of a total of 4 pixels, the average value of pixels 9 and 10 close to pixels 1 and 2 to be interpolated is discarded. The value of “pixel 3” may be used.
  • the frame image P f is usually the smallest rectangular area mainly composed of the object part, for example, the coding area (video obj ect plane) shown in FIG. 34A. Since the CA is divided into blocks (macro blocks) and encoded, the block located at the boundary of the encoding area must be divided into blocks (macro blocks). If the block adjacent to the upper left, upper left, or upper right corner of the block is located outside the signing area CA There is.
  • the pixel is located outside the code area CA. If the macroblock is a macroblock, it is closest to the pixel in the macroblock to which it belongs, as shown in Fig. 33, without referring to the reproduced binary value. It is convenient to assign the pixels directly to the blocks and use them.
  • the size of the data is smaller than that of the coding area CA in order to reduce the effects of errors.
  • the coding process may be closed on a per unit basis (this is called a "sync packet recovery unit (video packet)").
  • the “synchronous recovery unit” refers to each area surrounded by a dotted line indicated by a symbol “U n” in FIG. 34B, and the “synchronous recovery unit” is an encoding.
  • the area CA is an area that is divided into small areas, but it is still the case that it is composed of multiple macrob ports.
  • bilinear interpolation bi-linear
  • Interpolation using pixels included in the enlarged reference range wider than the reference range of (interpolation) reflects the inclination of the pixels in a wider range. And the problem of discontinuities is avoided. Also, the number of pixels used for interpolation is "9 pixels" rather than an even number such as "4 pixels”. If an odd number in the jar good of, the majority effect is obtained, et al. Are, Ru if the force s Oh discontinuity of avoidance effect is even more pronounced Ru obtained, et al. Is Ri good.
  • FIG. 26B shows an example in which interpolation is performed using 12 pixels.
  • the inside of a macroblock in a 1/2 reduced image of a certain macroblock is used. If the reproduced pixels at a certain position are P1, p2, p3, and ⁇ 4, and their immediate values (pixel values) are Ipi, Ip2, Ip3, and Ip4, These I 1, I p 2, I p 3, I p4 are
  • a is pixel A
  • b is pixel B
  • c is pixel C
  • d is pixel D
  • e is pixel E
  • f is pixel F
  • g is pixel G
  • h is pixel H
  • i is pixel I
  • J indicate a pixel J
  • k indicates a pixel K
  • 1 indicates a pixel L.
  • the “1/4” size conversion process may be implemented by performing the “1/2” size conversion process twice.
  • the technology disclosed by the present inventors in Japanese Patent Application No. 8-237053 includes a frame-by-frame (actually, a rectangular area containing an object).
  • An embodiment in which rate control is performed by applying size conversion and an embodiment in which rate control is performed by performing size conversion in small area units such as a block. It is presented. Further, in the above-described first to third embodiments, the rate control is performed by performing the size conversion in a more specific small area unit. showed that.
  • FIG. 35 is a diagram for explaining the file map encoding apparatus of the present embodiment.
  • This device has a reduction circuit 53 0, 52 1, 52 6, a binary image coding circuit 52 2, an enlargement circuit 52 3, 54 0 4, a motion compensation prediction circuit 52 5, multiplexing circuit
  • the binarized fan map image supplied via the file map signal input line 2 is a reduced circuit.
  • the data is reduced in frame units based on the size conversion rate CR.
  • Per frame Reduced signal is to via signal Line 5 0 2 is supplied to the A Le off ⁇ maps marks Goka circuits 5 2 0, after being divided into small areas, marks Goka of It is.
  • the alpha-map encoding circuit 520 is equivalent to the alpha-map encoding circuit 20 of FIG.
  • the components 52 1 to 52 6 of the up-coding circuit 52 0 are respectively the components 21 1 to 50 of the alpha-map coding circuit 20 in FIG. 12. Since it has the same function as 26, the explanation of the alpha-map encoding circuit 520 is omitted here.
  • the configuration of the alpha-map encoding circuit 20 of FIG. 12 is a simplified version of the configuration of the alpha-map encoding circuit 20 of FIG. Because of the representation, the configuration of the alpha-map encoding circuit 520 is equivalent to the alpha-map encoding circuit 20 in FIG. A configuration is also acceptable.
  • the size conversion rate CRb for each small area is multiplexed.
  • the signal is supplied to the multiplexing circuit 550 via the signal line 503, is multiplexed with the encoded information of the size conversion rate CR in frame units, and is multiplexed via the signal line 3. Output.
  • the reproduced image of the alpha-map encoding circuit 520 is supplied to an enlargement circuit 540 via a signal line 504, and the frame size is framed. After being enlarged based on the noise conversion rate CR, the signal is output via a signal line 4.
  • FIG. 36 is a diagram for explaining the decoding device of this embodiment.
  • This alpha-map decryption device uses a separation circuit 6 50, 643, binary image decoding circuit 641, reduction circuit
  • the encoded information supplied via the signal line 8 is divided into a size conversion rate CR in frame units by a demultiplexing circuit 6550, and It is separated into encoded information in small area units.
  • the coding information in small area units is supplied to an alpha-map decoding circuit 64 via a signal line 608, and the reproduced signal in small area units is converted to signal lines 60 through 80. It is supplied via 9 to an expansion circuit 660.
  • the description of the alpha-map decryption circuit 64 is omitted here because it is equivalent to the alpha-map decryption circuit 40 of FIG. .
  • the configuration of the alpha-map decoding circuit 40 in FIG. 13 is a simplified representation of the configuration of the alpha-map decoding circuit 40 in FIG. For this reason, the configuration of the alpha-map decoding circuit 640 may be equivalent to the configuration of the alpha-map decoding circuit 40 in FIG. Good.
  • the reproduction signal supplied via the signal line 609 is expanded based on the information of the frame-by-frame reduction ratio CR to expand the signal line 9. Output more.
  • the size of the alphabet map signal is converted in units of frames, and the size is converted in units of small regions.
  • size information such as size conversion rate information can be obtained. Reduced, which is particularly effective when encoding at lower encoding rates
  • FIG. 37 is a diagram showing an example of the resolution for each frame.
  • the frame at time n1 is resolved.
  • the playback must be stored in the frame memory without having to match the resolution to the resolution (in this case, the size conversion rate) of the frame at time n.
  • the image is converted to a frame-by-frame resolution (in this example, at time n, as shown in FIG. 37) by a file map encoding circuit 52.
  • the frame memory is used to store a playback image with a frame unit resolution supplied via the signal lines 504 and 609.
  • the frame memory in the latter stores the reproduced image of the original resolution supplied via the signal lines 4 and 9. . Therefore, when the frame memory is specified in the encoding device in FIG. 35 and the decoding device in FIG. 36, the former frame memory ( In the case of FM 1 (for the encoding device) and FM 3 (for the decoding device)), the result is as shown in Fig. 38 and Fig. 39.
  • the latter frame memory which will be referred to as FM2 (for the encoding device) and FM4 (for the decoding device)
  • Fig. 40 and Fig. looks like Figure 41.
  • the encoding device shown in FIG. 38 holds the information of the size conversion rate CR and the output information of the enlargement circuit 52 3 in a frame memory such as FM 1, and stores the information. It is configured to give the retained information to MC (Motion Compensation Prediction Circuit) 5 25.
  • the decoder shown in FIG. 39 is configured so that the information of CR and the expanded circuit 64 4 2
  • the output is stored in a frame memory such as FM3, and the stored output is configured to be fed to the compensation prediction circuit 6444.
  • the encoding device shown in FIG. 40 stores the information of the size conversion rate CR and the output information of the enlargement circuit 540 in a frame memory such as FM2. It is configured to give the retained information to the MC (Motion Compensation Prediction Circuit) 5 25.
  • the decoder shown in FIG. 41 uses the CR information and the output from the expansion circuit 66 0. It is configured to save the force and apply the saved output to the compensation prediction circuit 644.
  • FIG. 42A The specific structure of the frame memories FM1 and FM3 is shown in Fig. 42A, and the structure of the frame memories FM2 and FM4 is shown in Fig. 42A.
  • the specific configuration is shown in FIG. 42B.
  • the frame memories FM1 and FM3 include a frame memory mil for storing the image of the current frame and a frame memory mil for storing the image of the current frame. Size conversion ratio to which the retained image of mil is separately applied (size conversion process) A case conversion circuit (size conversion process) to process CR information conversion circuit) m 1 2, frame memory ml 3 that stores the output that has been subjected to size conversion processing in this size conversion circuit m 12, and , Frame memory F
  • M 2 and FM 4 have a frame memory m 21 for storing the image of the current frame and a frame memory m 21 for storing the image of the current frame.
  • the reduction image m 2 2 for reducing the retained image in accordance with the information of the separately provided size conversion rate CR, and the output reduced by this reduction circuit m 22 Save the frame memory ml 2 and force it.
  • frame memories FM 1 and FM 3 are connected to the frame by the signal line 504 and the signal line 609 in frame units.
  • the playback image of the corresponding resolution of the frame is supplied and stored in the frame memory ml1 for storing the current frame.
  • the frame memory ml 1 stores all the playback images of the current frame at the time when the encoding of the current frame (for example, time n) is completed. . 0 8
  • the size conversion circuit m 12 when the encoding of the frame at the time n + 1 is started, the frame at the time n is started.
  • the frame size conversion ratio CR at time n is
  • the reproduced image at the time n whose resolution has been converted by the size conversion circuit m 12 is stored in the frame memory ml 3 of the previous frame, and is stored at the time n + 1.
  • the encoding of the frame at the time n + 1 is started at the time n.
  • the playback image of the frame is loaded from the frame memory m 21, and the size of the frame at time n + 1 in frame units Perform reduction processing (resolution conversion) so that it matches the conversion rate CR.
  • the size conversion rate CR 1
  • the time n + in the example of FIGS. 37A to 37D. Since the size conversion ratio CR in 1 is, the resolution conversion process is not performed in the reduction circuit m 22. In the example of FIGS. 37A to 37D, the resolution of the frame at the time ⁇ + 1 at the current frame and the resolution at the time n + 2 is “1/2”. In the reduced circuit m 2 2, the size conversion rate CR is
  • the playback image of the frame at time n whose resolution has been converted by the reduction circuit m2 2 is the frame of the previous frame. It is stored in the frame memory m33 and serves as a reference image for the motion compensation prediction of the frame at time ⁇ + 1.
  • the frame memories F 1, FM 2, FM 3, and FM 4 describe the specific configuration and operation of the frame memories F ⁇ 1, FM 2, FM 3, and FM 4. It is characterized in that playback images with frame unit resolution supplied via signal line 504 and signal line 609 are accumulated.
  • the memories FM 2 and FM 4 are characterized by the ability to accumulate reconstructed images in frame units, which are supplied via signal lines 4 and 9, and have a power S feature. Because of that, each of them Various configurations are conceivable.
  • FIGS. 43A and 43B show an example of attribute information of a macroblock at time n and time n ⁇ 1.
  • the attribute information referred to here is
  • the tightest rectangle area containing the object part in the frame is placed in the rectangle, and the upper left corner of the rectangle is set so that it touches the boundary of the area.
  • the distribution (label distribution) of the attribute information of each mac block included in the set rectangular area is, for example, as shown in FIG. And 43B.
  • the label of the already encoded frames should be replaced.
  • Encoding the label of the current frame for coding can greatly improve the encoding efficiency, and in general, the encoding of the label at time n can be greatly improved.
  • the coding area (the smallest rectangular area mainly in the object part, for example, the coding area CA shown in FIG. 34A) in the The size of the coding area in the frame of the time n is different. In this case, as an example, in the procedure shown in FIGS. 44A and 44B, the size of the coding area at the frame at time n—1 is changed to the frame at time n. According to the size at.
  • the size of the encoded area in the frame at time n is greater than the size of the encoded area in the frame at time n-1. If the row is one row long and the column is one column short, as shown in Figure 44A, the coding area for the frame at the short time n-1 in the row is shown. At the right end of the area
  • the size of the coding area in the frame at time n — 1 is larger than the size of the coding area in the frame at time n. Is one column short, and a row is one line long In this case, cut the bottom one row of macroblock columns in the coding area, and then the rightmost one column of macroblocks in the coding area. Copy the row of row locks next to it and add one row.
  • FIG. 46A shows the difference between the attribute information of the above-described macroblock at time n and the attribute information of the above-described macroblock at time n_ 1 ′. The difference between each label at each mask block position and the result obtained by the same player at the same position are shown.
  • FIG. 46B shows the result obtained by calculating the difference between the label of the adjacent pixel position in the attribute information of the above-described macro block at time n.
  • the difference between the leftmost label and the label at the rightmost pixel position on one line is calculated, and the difference between the label at the leftmost pixel position and "0" is calculated. I decided to take it.
  • Fig. 46A is encoded between frames
  • Fig. 46B is encoded intraframe. 3 Call it rame coding) ::: and.
  • inter rr am e coding has a higher ratio of "S ,," than intra-frame coding, and inter-frame coding is more predictable.
  • the coding efficiency can be reduced compared to the intra-frame coding, although the amount of coding can be reduced.
  • a 1-bit code is used to perform intra-frame encoding or inter-frame encoding. Be switched so that the encoding is done by encoding in the frame. Since there is no label to refer to the frame to be encoded, the encoding within the frame must be performed, and the internal frame will be truncated. Replace No. need stomach
  • the prediction mode is, for example,
  • the “synchronous recovery unit” is, as described above, a subdivision of the rectangular area of the object, that is, a rectangular code. Area CA is added to the required This means that the data is divided in units of LOB blocks.For example, the code is divided or equal to each other so that the code amount of each synchronous recovery unit is equal. Group of macroblocks as a unit of synchronization recovery.
  • the reference block (the macro block to be referenced, adjacent to the macro block to which you belong) Macro block) Force S, even if
  • the switching may be performed for each "synchronous recovery unit", or may be performed for each frame, and may be performed for each sequence. People go even if they go.
  • the information on the mode in which the encoding was performed is sent from the encoding device to the decoding device.
  • the image has a general tendency. For example, as shown in Fig. 18, in the center of the frame, there is a probability that an object exists in the center of the frame. High, the probability that no object is present at the edge of the frame is high.
  • the macroblock that touches the edge of the frame uses a table that has a short code assigned to "transparent".
  • the code amount can be obtained even if prediction is not used. Can be reduced. This is the unpredictable mode.
  • This switching information is encoded, for example, in each "sync recovery unit", in a certain row, in each frame, and in each sequence.
  • FIGS. 47A and 47B are block diagrams showing the system configuration of the present embodiment for realizing the above-described processing, and these block diagrams are shown in FIGS. In the configuration of FIGS. 47A and 47B, the part surrounded by a broken line realizes the above-described processing in the configuration of FIGS. 47A and 47B. This is the part related to the example.
  • Fig. 47A shows an alpha-map coding device, which is an object area detection circuit 310, a blocking circuit 311, and a labeling circuit. Circuit 3 12, block coding circuit 3 13, label memory 3 14, size change circuit 3 15 5, label coding circuit 3 16, multiplexing Circuit (M UX) 3 17.
  • the object area detection circuit 310 is based on the input alpha-map signal and converts it to the alpha-map signal. And detects the rectangular area of the part containing the object, and provides information on the size of the rectangular area along with the information about the size of the rectangular area. Outputs a map signal.
  • the block circuit 311 1 is a circuit that blocks the alpha-map signal of this rectangular area at the mouth of the block, and is a labeling circuit.
  • the road 312 is connected to the macro block signal for each block of the macro block signal which is displayed by the macro block. Attributes (transparent only), Multi (mixture of transparent and opaque), and opaque (opaque only) of the contents of the phantom map signal in the The circuit to which the corresponding label ("0", "1", "3") is assigned.
  • the block encoding circuit 3 13 is a macro block having the label force S “1” (Multi) attribute, and the macro block in the macro block is displayed.
  • This is a circuit for encoding the alpha-map signal
  • the label memory 314 is a label information supplied from the labeling circuit 313.
  • the object area detection circuit 310 stores the size information of the area provided from the object memory via the label memory output line 302 and the label memory output line 302. This is memory for supplying the accumulated label information and size information to the size change circuit 315 together.
  • the size change circuit 3 15 is based on the label memory 3 14
  • the label information and size information of the frame supplied at time n-1 and the time n supplied from the object area detection circuit 310 are supplied.
  • This circuit changes the size of the frame so that the label information at time n-1 is equivalent to the size at time n, based on the frame size information and force.
  • the label-encoded circuit 316 uses the size-changed label information as a predicted value and supplies the label information supplied from the labeling circuit 312. Is a circuit that encodes
  • the multiplexing circuit 317 is composed of the coding information obtained from the label coding circuit 316 and the code supplied from the block coding circuit 313. This is a circuit that multiplexes the encoded information with the size information provided from the object area detection circuit 310 and outputs the result.
  • the alpha-map signal supplied via the signal line 301 is connected to an object area detection circuit.
  • the object area detection circuit 310 is provided to the rectangular area including the object from the alpha-map signal. Is detected. Information about the size of this rectangular area is output via a signal line 302, and the alpha-map signal inside the area is output to a block circuit 3. Supplied to 1 1.
  • the blocking circuit 311 converts the macroblock into a macroblock based on the alpha-map signal in this area.
  • the macro-blocked alpha-map signal is supplied to a labeling circuit 312 and a block signing circuit 313. 19
  • the attributes (transransparent, multi, opaque) of each macro block are determined, and the label corresponding to each attribute is determined. (“0”, “1,,,“ 3 ”).
  • the assigned label information is a block-encoded circuit. , Label memory 3 1 4 and label loop circuit 3
  • the label memory 314 includes the label information supplied from the labeling circuit 312 and the size of the area via the label memory output line 302. Information is accumulated, and the size change circuit is supplied to the size change circuit 3 15 from the label memory output line 303 together with the label information and the size information.
  • the label information supplied from the size change circuit 315 is used as a predicted value, and the labeling circuit 331 is used.
  • the label information supplied to 2 is encoded, and the encoded information is supplied to the multiplexing circuit 317.
  • the block encoding circuit 3 13 label Via the label memory output line 302 and the coding information supplied from the label coding circuit 3 13 and the label coding circuit 3 16. After multiplexing the supplied size information, the signal is output via a signal line 304.
  • the alpha-map encoder shown in FIG. 47B includes a demultiplexing circuit (DMUX) 320, a label decoding circuit 321, a size changing circuit 3 2 2, label memory 3 2 3, and block decoding circuit 3 2 4.
  • DMUX demultiplexing circuit
  • the demultiplexing circuit 320 is a circuit for separating the encoded information supplied via the signal line 305, and is a label decoding circuit.
  • the reference numeral 32 1 uses the information obtained by changing the size of the label information at the time n ⁇ 1 supplied from the size change circuit 3 2 2 as a predicted value, and uses the information at the time n as the predicted value. It is a circuit that reproduces bell information.
  • the size change circuit 3 22 2 is a circuit having the same function as the size change circuit 3 15, and is supplied from the label memory 3 2 3.
  • the label information and size information of the frame at time n—1 and the frame at time n provided separately from the demultiplexing circuit 320 are obtained.
  • the bell memory 3 2 3 is a circuit that operates in the same manner as the label memory 3 14, and is different from the label decryption circuit 3 2 1. Label information supplied after decryption and separation circuit
  • the block decoding circuit 32 which is a memory for supplying, is supplied from the label decoding circuit 32, and the regenerated lasers are supplied from the label decoding circuit 32. It is a circuit that reproduces the alpha-map signal for each block according to the bell information.
  • the demultiplexing circuit 320 separates the encoded information supplied via the signal line 305 into a block decoding circuit 3.
  • the size information via the signal line 306, while supplying it to the signal decoding circuit 324 and the label decoding circuit 321.
  • the label decoding circuit 321 the size of the label information in the frame at time n-1 supplied from the size changing circuit 3222 is changed. The information thus obtained is used as a predicted value, and the label information at the frame at time n is reproduced.
  • the reproduced label information is supplied to a block decoding circuit 32 4 and a label memory 32 3.
  • a block-by-block error is generated in accordance with the reproduced label information supplied from the labeling decoding circuit 3 21. Plays the format map signal.
  • the size change circuit 3 2 2 is the size change circuit 3 15 and the label /
  • one frame in a frame is adjacent to one block (one block) between the frames.
  • the VLC (Variable Length Coding) table was switched with reference to the lock. In this case, if the inter-frame correlation is high, refer to “Adjacent blocks between frames”. If the inter-frame correlation is low, “ The VLC table was switched with reference to the "adjacent block in the frame”. However, in practical applications, it is often better to use both inter-frame correlation and intra-frame correlation.
  • the mode of a certain pixel position is set to “M (h, v, t)” (where h, v, and t represent horizontal, vertical, and time-direction coordinate axes, respectively).
  • M (x, y, n) for example, "M (X — 1, y, n),,, “M (x, y-1, n)” and "M (x, y, n-1)" to select the VLC table.
  • the number of modes is three as shown in FIGS. 43A and 43B, the number of reference blocks is three blocks (three macro blocks). ).
  • the number of VLC tables is
  • This section describes a method of encoding the attribute information of a book using the label of the previous frame for prediction.
  • FIG. 48 is a block diagram of an encoding circuit as one embodiment of the present invention.
  • this encoding circuit is composed of an object area detecting circuit 720, a block circuit 704, a labeling circuit 706, It consists of a label encoding circuit 708, a label memory 709, a reference block determination circuit 710, and a prediction circuit 712.
  • the object area detection circuit 702 is a block containing the object of the Alphamap signal 701 and the object containing the object.
  • a region represented by a multiple of the size is set as a coding region, and at the same time, the alpha map signal 703 in the coding region is cut out.
  • the block-forming circuit 704 uses the extracted alpha-map signal 703 as a block unit of 16 ⁇ 16 pixel configuration.
  • the output is divided (blocked) into blocks (macroblock units), and the labeling circuit 706 is blocked.
  • a predetermined label is assigned to the alpha-map signal 705 according to the content of the object and output as label information 707 It is something.
  • the label encoding circuit 708 switches the encoding table according to the given prediction value 714 to encode the label information 707.
  • the label memory 709 is provided with the label information 7 assigned to each block by the labeling circuit 706. 07, and the reference block determination circuit 71 0 is located at the same position as the encoding block in the previous frame.
  • the reference processing is performed when the block is determined as the reference block 711, and the prediction circuit 712 is held in the label memory 709. Refer to the label 711 of the previous frame that has been set, and predict the label at the position of the reference block 711 and label it as the predicted value 714.
  • Bel coding circuit 7 0 It is sent to 8.
  • the signal 7001 of the alphabet map is input to the object area detection circuit 702.
  • an area including the object and represented by a multiple of the block size is set as an encoding area.
  • the output of the alphabet map 703 cut out in the coding area is sent to the S block circuit 704.
  • the block circuit 704 is divided into block units (macro block units) of a pixel map S 703 pixel structure S 16 X 16 pixel configuration.
  • the blocked fan-out map 705 is a labeling circuit 70
  • Label information 707 (mode information) is added for each block. Label information 707 is sent to a label encoding circuit 708 and stored in a label memory 709. The label memory 709 stores the previously encoded label.
  • the reference block determination circuit 71 for example, the block force at the same position as the coding block in the previous frame; Lock 711 is determined as Route 7 12
  • the label 713 of the previous frame is also input from the label memory 709, and the position of the reference block 711 of that frame is also input.
  • the label encoding circuit 708 the encoding table is switched according to the predicted value 714 to encode the label information 707, and the code 715 is used.
  • the force S is output.
  • the coding area is always equal to the frame, and sometimes the reference block is determined to be one.
  • the coding area is smaller than the power S frame, and the position of the coding area differs between the previous frame and the current frame.
  • the reference block differs depending on whether to use a coordinate axis whose origin is the corner of the frame or a coordinate axis whose origin is the corner of the coding area.
  • Figs. 49A and 49B show the frame images Fn-1 and Fn at time n_1 and time n, respectively, and their respective frames Fn-1 and Fn. This is an example of the mode information MD of each macroblock of the coding area CA in n.
  • the left end and the top end of the coding area change.
  • the blocks corresponding to the mode information of the current frame are indicated by shaded portions as shown in FIGS. 50A and 50B. Since there is no mismatch (21), encoding using this value may reduce the encoding efficiency.
  • the origin Fc0 of the current frame and the origin Fp0 of the previous frame are made to coincide with each other on the coordinate axes of the frame.
  • the block at the nearest block position is referred to as a block.
  • the block looks like Figure 50B. That is, in the example of FIGS. 49A and 49B, since the left end and the upper end change, the left end and the upper end are “cut” or “no update” as shown in FIG. 50B. "This will match the size of the coding area of the current frame. In this case, the block corresponding to the mode information of the current frame is, as shown in FIG. 50B, the unmatched block or block. Only the shaded blocks (three) are available.
  • the label of the previous frame is encoded according to the situation.
  • the coding efficiency can be improved by switching between changing based on the coordinate axes of the area or changing based on the coordinate axes of the frame. It can be achieved. Coordinate axes can be determined by selecting the most appropriate one on the encoding device side and sending the switching information, or by using known information for both the encoding device and the decoding device. You may decide on this.
  • FIGS. 45A and 45B are examples in which it is better to use the coordinate axes of the coding area.
  • the frame Fn-1 in Fig. 45A and the frame Fn in Fig. 45B are used.
  • Frame; ⁇ shows that it has changed greatly.
  • the position of the coded area in the frame is greatly shifted as can be seen from the drawing. Determining the reference block is not a good idea.
  • the current frame (frame Fn at time n) and the previous frame (frame Fn-1 at time n-1) are within the frame. If the position of the coding area CA of the coding area shifts greatly, it is better to use the coordinate axes of the coding area, and the position of the coding area in the frame will be more accurate. If not, the force S using the frame's coordinate axes is used.
  • the encoded data includes the layer of the coding area, the layer of the macro block MB, and the layer of the linear area, and the layer of the coding area.
  • the information includes coding area size information, coding area position information, coding area size conversion rate information, and the like.
  • the MB layer consists of binary shape information, texture MV information, multi-valued shape information, and texture information.
  • the binary shape information is mode information, motion vector information, and size. Includes conversion rate information, scan direction information, and binary coding information.
  • the coding area size information indicates information indicating the size (vertical and horizontal size) of the coding area
  • the coding area position information indicates the coding area.
  • the coding area size conversion rate information is the size conversion rate (CR) of the binary image in coding area units. Refers to information.
  • the MB encoded information indicates information for reproducing the object in the MB.
  • the binary shape information in the MB layer refers to information indicating whether each pixel in the MB is inside the object, and the texture MV information is the luminance signal or color difference in the MB. It refers to the motion vector information for predicting the motion of the signal and the multivalued shape information is the weight when combining an object with other objects. Weighting information, and the textual information is the luminance signal and color in the MB. 30 Refers to the encoding information of the difference signal.
  • the mode information in the binary shape layer indicates information indicating the attributes of the binary image in the MB
  • the moving vector information indicates the binary image in the MB. It refers to the motion vector information for motion compensation prediction
  • the size conversion rate information refers to the size conversion rate (CR) information of the binary image in MB.
  • the canon direction information indicates the information indicating whether the encoding order is the horizontal direction or the vertical direction
  • the binary encoding information indicates the encoding information of the binary image.
  • the difference between "rev ref and curr ref” is large, and the sign is used for the current frame and the previous frame.
  • the size of the coding area does not substantially change, it is a component of the better rate based on the coordinate axes of the coding area.
  • the information of "prev-ref and curr_ref" and the size of the coding area are coded prior to coding of the coding area. Since both sides have known information, both coordinate axes must be No additional information is needed to identify whether it has been used.
  • the outside of the coding area 731 is The label is undefined.
  • the outside of the coding area 731 may be a reference block. You will need to insert the label for.
  • FIG. 52D is an example in which a predetermined value, in this case, “0” is inserted in the undetermined label portion in FIG. 52A.
  • Fig. 52C is an example of the undefined label in Fig. 52A, which is outside the coding area. In the previous frame, the next frame was added to the part where the object force was weak, for example, the movement was large and the shape changed drastically. This is useful when the probability that an object will appear is high.
  • FIG. 52B shows the undefined part of the label in FIG. 52A.
  • the coding area of the next frame is shown. Is extrapolated, and the other parts are overwritten, for example, in this case, the frame before the frame two or more frames earlier You can use the labels to make predictions.
  • the block that is the target block of the current frame to be reduced and obtained for example, the block of the previous frame corresponding to the macro block MB1.
  • the masks ⁇ -blocks MB 2, MB 3, MB 4 and MB 5 of the previous frame become the block openings MB 1 after reduction.
  • ⁇ MB5 The address of ⁇ MB5 is given by (2X, 2y, (2 + 1, 2y), (2X, 2y + 1), (2x + l, 2y + 1). .
  • the coefficient "2" in is given as the ratio of the value of the size conversion ratio CR between the previous frame and the current frame.
  • the simplest and least computationally expensive method is to use the label of a block located at one of four predetermined positions (for example, upper left). .
  • One label has the same four labels When there is something, if the label with the largest number is used as the predicted value, the probability of achieving the predictive power S is increased.
  • the labels appear in the order of appearance frequency in descending order. , And the higher-ranked label in that order is selected as the predicted value.
  • a rectangular coding area CA is set in a frame, the coordinate axes originating from the corners of the coding area CA are used, as shown in Fig. 54.
  • the boundary of the block of the frame (the macro block to be encoded) overlaps the boundary of the macro block of the reference frame
  • the area to be coded can be positioned in steps smaller than the width of the macrobloc, and the origin is the frame corner.
  • the coordinate axes are used, as shown in Fig. 55, the boundaries of the blocks generally do not overlap, and a total of nine macroblocks, MB6 to MB14, are referred to. Will be illuminated.
  • the lock MB 19 refers to the lower right portion of the macro block MB 15 in the reduced frame.
  • the powerful block that has the label of block MB15 as the predicted value is the reference block MB16.
  • Figure 57 shows a block diagram of a configuration example of the decoding device of the present invention in which the labels are used for prediction.
  • This decryption device includes a label decryption circuit 716, a label memory 717, a reference block determination circuit 718, and a prediction circuit 720. It is composed of
  • the label decryption circuit 711 decrypts the label data from the input encoded data to be decoded.
  • the label memory 511 stores the data of the decoded label
  • the reference block determination circuit 718 stores the data of the decoded label.
  • the block located at the same position as the sign block is determined as the reference block 711.
  • the prediction circuit 720 calculates the prediction value 722 from the label 721 of the previous frame and the reference block 719. Therefore, it has a function given to the label decoding circuit 7 16.
  • the stream 715 of the coded data to be decoded is input to the label decoding circuit 716. And the label is decrypted.
  • the labels that have been decrypted so far are stored.
  • the reference block determination circuit 711 the reference block 711 is determined in the same manner as described for the encoding device, and the prediction block 711 is determined.
  • Rd 7 2 Sent to 0.
  • the predicted value 722 is obtained from the previous frame's label 721 and the reference block 711, and the label is obtained. It is sent to the decryption circuit 7 16.
  • the predicted value 716 is sent to the decryption circuit 7 16.
  • the mode of a certain block “M (h, V, t), ..., where h, V, and t represent horizontal, vertical, and time-direction coordinate axes, respectively.
  • h, V, and t represent horizontal, vertical, and time-direction coordinate axes, respectively.
  • the mode used here is a part of the information of the motion vector used for motion compensation. And the following set of modes (this is called the mode set A).
  • (3) and (4) in mode set A are both coy modes, but (3) in mode set A is dynamic. This means that the vector is zero, and (4) in mode set A means that the moving vector is other than zero.
  • mode set A (4) it is necessary to separately encode the value of the moving vector, but in the case of mode set A (3), the moving vector is required. It is not necessary to encode the file.
  • the probability of being a motion vector is high, if the mode set A is used, the code amount of the mode and the code amount of the motion vector Can be reduced.
  • Step A5 "M (x, y, n) ,
  • Step A6 "M (X, y, n)" is encoded using the encoding table of mode set A. End of encoding
  • the decoding process uses the same algorithm as the flowchart in Fig. 58 to convert the encoding table for mode set A or mode set. Decide which one to use for B, and decrypt using that table.
  • FIG. 59 shows another algorithm that can obtain the same effect as the above-mentioned algorithm. here .
  • Step B4 Refer to the "M (*, *, *)" force. In the case of “transparent”, select "M (*, *, *)".
  • Step B5 “M (X, y, n) ,, is a mode set
  • Step B6 Encode "M (x, y, n)" using the encoding table of mode set C. End of encoding.
  • Step B7 "M (x, y, n) , is encoded using the encoding table of mode set D.
  • the encoding is completed.
  • Block attributes include encoding parameters, such as block size and block reduction ratio, and encoding scan.
  • the direction and the value of the motion vector can be included as needed.
  • the reference block of the previous frame is used for the prediction, but this reference block is used.
  • the alpha map itself may be used for prediction. That is, the alpha map is recorded in the memory, and each time the attribute of each block is encoded, the attribute of the reference block is encoded.
  • the block used when encoding the previous frame is the reference block whose position is shifted by several pixels. You can do it with force S. This means that the reference block does not overlap with the block used when the previous frame was encoded, and it does not overlap if it does not fit in with the pitch. Thus, a more accurate prediction can be made.
  • the encoding table is switched using the label block of the reference block.
  • the background is divided into a background and an object, and the encoding is performed by dividing the screen into a background and an object.
  • the encoding is performed by dividing the screen into a background and an object.
  • the alpha-map signal is encoded together with the encoded information of the image to form a bit stream, which is then transmitted.
  • the former is used for broadcasting and personal computer communication, and the latter is used for merchandise containing content such as music CDs. Will be traded.
  • the encoding information of the image and the compression code of this alpha-map signal are stored on the storage medium. It accumulates in a bitstream that combines what has been converted into a single medium, and stores the contents over a long period of time on a single medium to watch movies and the like. It is intended for storage media storing compressed and encoded bitstreams containing such alpha-map images.
  • FIG. FIG. 60 shows the VLC multiplexed by the multiplexing circuit 180 in FIG. 19, the mode information (shape mode) b 0 N, the motion vector information b 1, Conversion rate information
  • the binary image encoding information b4 indicates the format (ij) of the symbol (J (bitstreara)), and in this invention, the binary image
  • the information from b0 to b3 be decoded, and that the mode information b0 is all other information.
  • the information b 1 to b 4 in If it has not been decrypted before, no other information can be decrypted. Therefore, each of the information b0 to b4 must have a configuration in which mode information is arranged at the beginning and binary plane image encoding information is arranged at the end as shown in FIG. No.
  • FIG. 61 is a diagram illustrating a system that reproduces an image signal using the recording medium 810 in which the code string illustrated in FIG. 60 is stored.
  • a code sequence including the code sequence shown in FIG. 60 is stored in the recording medium 8100.
  • the decoder 820 reproduces an image signal from the code string stored in the storage medium 810.
  • the image information output device 830 outputs a reproduced image.
  • FIG. 60 is a code string of a format as shown in FIG. 60.
  • the decoder 830 reproduces an image signal from the code string stored in the storage medium 810. That is, the decoder 820 reads the code string from the storage medium 810 via the signal line 8101, and performs the procedure shown in FIGS. 62 and 63. To generate a playback image.
  • FIG. 63 is a flowchart of the step (S5) of “2D direct image playback” in FIG. 62.
  • step S1 the mode information is restored (step S1), and the restored mode! Checks whether any of the following is applicable: "transparent information, opaaue", or "no update”. Steps S2, S3, S4).
  • step S7 If it is "oaqe”, all pixels in the macro block are made opaque and the processing is terminated (step S7). Also, transparent and t> ⁇ , o If it is not "aque” but "no u date", then the motion vector will be restored to the 'If report (step S8), and the motion compensation will be predicted (step S8). In step S9), the inside of the macro block is copied with the obtained motion compensation predicted value (step S10), and the process ends.
  • step S5 If it is neither transparent nor ⁇ aque, and it is not "no update", the process proceeds to the binary image reproduction process (step S5).
  • the processing in S5 is as shown in FIG. 63.
  • it is checked whether or not the coding is "inter” (step S21). As a result, if the coding is "inter”, the motion vector information is decoded (step S25), the motion compensation prediction is performed (step S26), and the next Then, the size conversion rate information is decoded (step S22), and the scan direction information is decoded (step S23). The binary encoded information is decoded (step S24), and the process ends.
  • step S21 "inter" If the encoding is not performed, the size conversion rate information is decoded (step S22), and the scan direction information is decoded (step S2). 3). The binary encoded information is decoded (step S24), and the process ends.
  • the decoder 8220 reproduces an image, and supplies the reproduced image to the image information output device 830 via the signal line 802. As a result, the reproduction plane image is displayed on the image information output device 830.
  • the image information output device refers to, for example, a display printer or the like.
  • the image information output device refers to, for example, a display printer or the like.
  • the sign of the frame unit size conversion rate is placed before all the small area code strings in the frame.
  • a moving image as content and an alpha-map signal are compression-encoded and stored as a bit stream in the storage medium.
  • a reproduction system of the storage medium can be provided.
  • FIG. 19 and FIG. 20 are diagrams showing the framework of the present invention.
  • the motion vector detection The motion vector detected by the circuit 178 is supplied to the MV encoding circuit 179 via the signal line 107, and after being encoded, the motion vector is detected by the VLC.
  • the signal is supplied to a multiplexing circuit 180, multiplexed with other encoded information, and output via a line 3.
  • FIG. 20 the motion vector information separated from the encoded information supplied via the signal line 8 by the VLD 'separation circuit 210 is shown.
  • b 1 is reproduced as a moving vector signal by a moving vector reproducing circuit 290.
  • the present embodiment relates to the MV encoding circuit 179 and the moving vector reproduction circuit 290 described above.
  • the correlation of the motion vector signal is strong between adjacent blocks, and therefore, the motion vector signal must be corrected by predictive coding to remove the correlation. Is encoded.
  • FIG. 64 is a diagram for explaining an example of predictive coding of a motion vector.
  • each rectangular frame shows a macro block, and the base is a dot.
  • the rectangular frame indicated by the turn is the block to be encoded.
  • the predicted vector MVP s for the MV s of the block to be encoded is calculated. To do this, use the macro block immediately before the block to be coded (the block next to the block to be coded in Figure 64).
  • the motion vector MV s1 and the macro block immediately above the block to be coded (the block to be coded in FIG. 64). Movement of the block (above) MV s2, and then use the motion vector MV s of the block to the right of the macro block immediately above.
  • the predicted vector MVPs for the motion vector MVs of the block to be encoded is generally obtained by calculating the prediction vector MVPs around the block to be encoded.
  • the motion of the block can be obtained using the MVs1, MVs2, and MVs3.
  • the horizontal and vertical components MVPs—h and MVPsV of MVPs are obtained as follows.
  • M V P s h Median (M V s 1 ⁇ h, M V s
  • the motion of each block in the I injection of MVs1, MVs2, and MVs3 is described.
  • Figures 65A and 65B show the frame images F n — 1 and F n at time n — 1 and time n, respectively, and the mark at each frame.
  • Coding area CA n —:! , CA n If the block around the block to be encoded is a block that does not include an object, or if the block is a block that does not include an object, the block should be moved to that block. The vector does not exist. Also, in the case of an intra-frame coded block, there is no moving vector S in the block.
  • the default value is used for the predicted vector MVPs. It will be.
  • the default values are the vector “prev—ref” shown in FIG. 65A and the vector “curr” shown in FIG. 65B.
  • the feature is that the difference vector of "refset” with “of f set” and “zero vector” are adaptively switched and used.
  • Switching between the default setting "of f set” and “zero vector setting” is, for example, the time n obtained based on the frame coordinate axes. Based on the error value of the object at time n — 1 and the coordinate axis of the coding area The error value of the object at time n and the error value of the object at time n — 1 are compared. If the former is large, “of f set” is added. If the latter is large, "zero vector" should be used as the default value.
  • FIGS. 66A and 66B show examples of the implementation of the MV encoding circuit 1779 and its associated circuits in the system shown in FIG.
  • FIG. 66A is a lock value diagram
  • FIG. 66A is a default value calculation circuit as an ancillary circuit
  • FIG. 66B is an MV coding circuit 179.
  • the default value calculation circuit as an ancillary circuit shown in Fig. 66A is used for processing the current processing frame and the previous processing frame for the object area. From the frame, the position movement of the object of the frame as viewed from the previous point in time is calculated as the movement vector value between the frames. As shown in FIG. 66A, the encoding area detection circuit 910, the default value determination circuit 911, and the area It is composed of an information memory 912, an offset calculation output circuit 913, and a selector 914.
  • the signal line 902 is a signal line for inputting the current frame data, and corresponds to the signal line 2 of the system in FIG. 19. The purpose is to receive the current frame data input from signal line 2 as input.
  • the signal line 902 in FIG. 66A is a signal line for supplying the frame data at the previous time held in the frame memory 130. Therefore, the frame data at the previous time is supplied from the frame memory 130 via the signal line 902.
  • the signal line 903 is a signal line for outputting flag information from the default value determining circuit 911, and the signal line 904 is an encoded signal. This is a signal line for supplying information on the coding area CAn from the area detection circuit 910, and the signal line 906 is read from the area information memory 912. This is a signal line that supplies the position information of the coding area CA n — 1.
  • the coding area detection circuit 910 is configured to perform coding based on the image signal of the current frame Fn-1 supplied via the signal line 901. It detects the size and position information VC0 of the generalized area CAn, and the detection result is sent to the default value determination circuit via the signal line 904.
  • the circuit 911 is supplied to the area 911, the area information memory 912, and the offset calculation output circuit 913.
  • the area information memory 912 includes the coding area CA n This memory stores the size and position information of -1. After the frame encoding at the time n is completed, the memory of the encoding area CA n is stored. This will accumulate noise and location information.
  • the offset calculation output circuit 913 includes a coding area CAn supplied via a signal line e4 and a coding area supplied via a signal line 906. Using the location information of CA n — 1, the vector value "of f set" is obtained and given to the selector 9144.
  • the selector 914 is provided with “zero vector” which is a moving vector value of zero and an offset calculation output circuit 913. Is input, and either one of them is supplied from the default value determination circuit 911 via the signal line 903. This circuit is selected according to the flag, and the vector value selected by the selector 914 is set as a default value via the signal line 905. Then, it is a circuit that outputs to the selector 923 of the MV encoding circuit 1779.
  • the MV coding circuit 179 has the following characteristics, as shown in FIG.
  • MV memory 921 It is composed of MV memory 921, MV prediction circuit 922, selector 923, and difference circuit 922.
  • the MV memory 921 is a dynamic vector detection circuit 178 supplied via the signal line 107 shown in FIG. In the memory that saves motion vector information Yes, the moving vector MV sn (n-1, 2, 3) around the block to be encoded is held in the MV memory 179.
  • the MV prediction circuit 922 is composed of a motion vector MV sn (n ⁇ 1, 2, 3) around the block to be encoded supplied from the MV memory 921. ) Is a circuit for obtaining the predicted vector MVPs.
  • MV sn n—1, 2, 3
  • the prediction vector MVP s cannot be obtained normally, so that the MV prediction circuit 1 ⁇ 92 has a function to output a signal to identify whether or not the predicted vector MVPs has been normally obtained, and the identification signal is sent to the signal line.
  • a mechanism is provided to supply the selector 923 via the 925.
  • the selector 922 receives the MVPs supplied from the MV prediction circuit 922 and the default value given via the signal line 905 as an input. Then, according to a signal supplied via the signal line 925, one of the two is selected and supplied to the differential circuit 924.
  • the differential circuit 922 4 is a circuit for obtaining a prediction error signal of the motion vector, and is a motion vector detection signal supplied via a signal line 107.
  • the image signal of the frame Fn is input to the default value determination circuit 911, and the image signal of the frame Fn-1 is set to the default.
  • the value is input to the value determination circuit 911 and the coding area detection circuit 910.
  • the size and position information VC0 of the coding area CAn are detected based on the image signal of the frame Fn_l.
  • the detection result is sent via a signal line 904 to a default value determination circuit 911, a region information memory 911 and an offset. It is supplied to the calculation circuit 913.
  • the encoding area CA n from the encoding area detection circuit 910 supplied via the signal line 9104 is output.
  • the error amount between the frame Fn and the frame Fn_l is compared with the error amount between the coding areas CAn and CAn-1. If the result of the comparison is that the former is large, the default value Use "of f set” as the default value, otherwise use zero vector as the default value, and use the default value.
  • Use "of f set" as the flag output the flag information to identify whether to use zero vector, or output via signal line 903. .
  • the flag value output from the default value determination circuit 911 via the signal line 903 is output to the coding area detection circuit via the signal line 904.
  • the code in the data format shown in FIG. 51 together with the information on the size and position of the coding area CA n output from the circuit 910 is used. Multiplexed to the layer of the coding area. Then, the data is transmitted or stored in a recording medium.
  • the area information memory 912 is a memory for storing the size and position information of the coding area CAn-11, and after the encoding at time n is completed, the area information memory 912 stores the code. It stores the size and position information of the coding area CA n.
  • a coding area CAn supplied via a signal line 904 and a code supplied via a signal line 906 are used. After obtaining the vector value "of f set" using the position information of the coding area CA n — 1, it is supplied to the selector 914.
  • the selector 914 according to the flag supplied from the default value determination circuit 911 via the signal line 903, "off Either "set” or a zero vector is selected and this is the default value.
  • This default value is the signal
  • the signal is output via a line 905 to a selector 923 of the MV encoding circuit 1779.
  • the motion vector MVs of the block to be encoded is supplied to the MV encoding circuit 179 of FIG. 66B via the signal line 107, and the MV encoding circuit 179 is provided with the MV encoding circuit. It is supplied to the memory 921 and the difference circuit 924.
  • the MV prediction circuit 922 includes, in the MV memory 921, a motion vector around the block to be encoded from the MV memory 921 MV sn (n—1, 2, 3, 3); ⁇ Supplied and the predicted vector ⁇ V ⁇ s Power S is obtained.
  • MV sn n—1, 2, 3, 3
  • MV sn n—1, 2, 3
  • ⁇ Supplied and the predicted vector ⁇ V ⁇ s Power S is obtained.
  • the predicted vector MVP s is not normally obtained, and the predicted vector MVP s is not obtained.
  • a prediction error signal of the motion vector is calculated, and the calculation result is used as the motion vector information b 1 as the MV encoding circuit 17.
  • FIGS. 67A and 67B are block diagrams showing an embodiment of a decoding circuit for realizing the present invention, in which MV-encoded data is shown.
  • FIG. 2 is a block diagram showing a main part configuration for reproducing the MV.
  • the MV reproducing circuit 290 and its accompanying devices in the system shown in FIG. 20 are shown.
  • FIG. 3 is a block diagram showing an embodiment of a circuit.
  • FIG. 67A shows a default value operation circuit as an auxiliary circuit
  • FIG. 67B shows an MV reproduction circuit 290.
  • the default value calculation circuit as the auxiliary circuit shown in Fig. 67A includes the area information memory 101, the offset calculation output circuit 101, and the cell. It is composed of a cluster and a force. Also, 1001 is included in the upper layer of encoded data transmitted or data stored in a storage medium and read out. A signal line that gives information on the flag for identification of default value selection, and 1002 is included in the upper layer of data that has been encoded and transmitted. This is a signal line that gives positional information on the area CA n that is being encoded, and corresponds to 1003 and 1004 on the coded circuit side.
  • 1003 is a signal line for giving position information of the area C ⁇ n-1 and 1004 is a signal line for outputting a default value.
  • the area information memory 11010 is a memory for storing the position information of the area CAn-1.
  • the offset calculation circuit 1011 is provided with a signal line 1011. Location information of the area CA n supplied via the line 0 2, and supplied via the signal line 100 3 The vector value "off set” is obtained using the location information of the area CA n — 1 to be obtained, and the vector value "of f set "Is configured to be supplied to the selector 101.
  • the selector 1012 should be given to the predetermined moving vector value of zero and the offset calculation circuit 1011.
  • One of the two types of "of f set" is the information handling of the default value selection and identification flag given via the signal line 1001.
  • the output of the selector 1102 is output as a default value to the signal line 104, and the MV is played back. This is the configuration given to circuit 290.
  • the above is the configuration of the default value calculation circuit on the decoding side.
  • the MV reproduction circuit 111 has a calorie calculation circuit 1101, a selector 1102, an MV prediction circuit 1103, and an MV memory. 1 1 0 4 and power.
  • Arithmetic circuit 1101 is a motion vector information b 1 which is a prediction error signal of the motion vector of the block to be decoded. And a default value given via the selector 1102, and the two are added together to output the result.
  • the output is configured to be output to the MV memory 1104 and the signal line 203 in the configuration of FIG.
  • the configuration is such that the power is supplied to the selector 1102 via the 1105.
  • the selector 1102 has a default value provided via the signal line 104 and a prediction value provided from the MV prediction circuit 1103.
  • Vector MVPs are input, and one of them is selected and added according to the identification signal supplied via the signal line 1105, and the addition circuit 1101 is selected. This is the circuit given to
  • the signal line 1001 has a default value of "of f set", and a signal line using zero vector. Flag is supplied to identify the signal line 1 1 0 2 Is supplied with the location information of the area CA n.
  • the position information of the area CAn supplied via the signal line 1002 is supplied to the area information memory 1010 and the offset calculation output circuit 1011. .
  • the area information memory 1010 stores the position information of the area CA n — 1, and stores the position information of the area CA n after the end of the time n decoding. This will be.
  • the position information of the area CAn supplied via the signal line d2 and the position information supplied via the signal line 103 are supplied. After obtaining the vector "of f set” using the location information of the area CA n — 1 and supplying the vector value "of f set" to the selector 1102 You
  • the “motion vector”, which is the prediction error signal of the motion vector of the block to be decoded, is used. Is supplied to the addition circuit 1101.
  • MV sn (n 1, When (2, 3) does not exist, the predicted vector VP VP s was not normally obtained, and thus the predicted vector ⁇ VP s was obtained normally.
  • the signal for identifying whether or not the signal is supplied to the selector 1102 via the signal line 1105.
  • the signal is supplied from the ⁇ V prediction circuit 1103 according to the identification signal supplied via the signal line 1105.
  • ⁇ VP s key or the level of the default value which is the output of signal line 104 is selected and supplied to the power calculation circuit 1101 .
  • the motion vector prediction error signal (“motion vector information b1”) and the prediction signal MVPs are added, and thus, the motion vector MVs of the block to be decoded is reproduced.
  • the motion vector MVs of the block to be decoded is output from the MV regeneration circuit 110 through the signal line 203 and the MV is output. It is stored in memory 1104.
  • the alpha image which is sub-image information indicating the shape of an object, a position in a screen, and the like. It is now possible to efficiently encode the information of the zipper and to decode it. An image coding device and a decoding device can be obtained.
  • the encoding amount of the alphabet map can be reduced, the conventional encoding in which the encoding is performed in frame units is performed. Compared to the method, it is possible to encode each object separately without a significant decrease in encoding efficiency.
  • the encoding amount of the alphabet map can be reduced, so that the encoding method is compared with the conventional encoding method in which encoding is performed on a frame basis.
  • the encoding can be performed separately for each object without a large decrease in encoding efficiency.

Description

明 細 書 画像符 号化装置お よ び画像復号化装置 技術分野
本発 明 は 、 画像信号 (video )を 高能率 に符 号化 し 、 伝送 · 蓄積す る と 共 に 、 復 号す る た め の 画像符 号化装 置 (video encoding apparatus )お よ び画像復 号ィ匕 装置 video decodina apparatus ) に 関す る 。
背景技術
画像信 号は膨大 な 情報量 を持っ た め 、 伝送や蓄積 に 供す る 場合 に は圧縮符 号ィ匕 ( compress ion-encode ) す る の が 一般的 で あ る 。 画像信 号 を 高 能率 に符 号化す る に は 、 フ レ ー ム 単位 の 画像 を 、 所要画素数 単位 (例 え ば、 M X N 画素 ( M : 水 平方 向 の 画素数 、 N : 垂直 方 向 の 画素数) ) で ブ ロ ッ ク に 分割 し 、 そ の 分割 し た 各 プ ロ ッ ク 毎 に 直交変換 し て 画像 の持つ 空 間 周 波数
( spatial f requency ) を 各周 波数成分 に 分離 し 、 変 換係数 と し て 取得 し て こ れ を 符 号化す る 。
と こ ろ で 、 画像符 号化 の 一つ と し て 、
U J . Y . A . Wang e t . a 1 . Applying Mid-level Vis ion Techniques f or Video Data
Compres s ion and Manipulation " , M . I . T . Media Lab . Tec h . Report No . 263 , Feb . 1994 ,』 に お レヽ て 、 ミ ッ ド レ ベ ル符 号化 と 呼 ばれ る 範疇 に 属す る 画像符 号 化法が提案 さ れて い る 。
こ の方式で は 、 背景 と 被写体 (以後 、 被写体 を ォブ ジ - タ ト と 呼ぶ) カゝ ら な る 画像 が あ つ た と し て 、 こ の 背景 と オブ ジ ェ ク ト を 分 け て符 号化 し て い る 。
こ の よ う に 、 背景やオブ ジ ェ ク ト を 別 々 に符 号化す る た め に は 、 オブ ジ ェ ク ト の 形状や画面 内 の位置 を表 す例 え ば 2 値の 副 画像情報 ( subs idiary video inf ormat ion )で あ る ァ ノレ フ ァ マ ッ プィ言 号 ( a lpha- map s i g n a 1 ) 力 必要 と な る 。 な お 、 背 景 の ア ル フ ァ マ ッ プ信 号は 、 ォブ ジ ェ ク ト の ア ル フ ァ マ ッ プ信 号か ら 一意 に求 め ら れ る 。
と こ ろ で 、 こ の ア ル フ ァ マ ッ プ信号 を 効率的 に符 号 化す る 方法 と し て 、 2 値画像 の符 号化法 (例 え ば、 M M R ( Modi f ied Modi f ied READ) 符 号ィヒ等) や 、 線図 形 の 符号化法 ( チ ェ イ ン符 号化 ( c ha in
encoding )等) 力 用 レヽ ら れて レヽ る 。
ま た 、 更 に ア ル フ ァ マ ッ プの符 号量 を 低減す る た め に 、 形状 の 輪郭線が ポ リ ゴ ン ( p o 1 y g o n s ) に よ っ て 近 似 さ れ ( approx ima t e ) ス プ ラ ィ ン 曲線 ( spl ine curves ) で ス ム一シ ン グ さ れ る 方 法 ( j . Os termann ,
Ob j ec t -based analys i s - synthe s i s c o d i η σ bas ed on the s ource model o f moving ri gid 3 D ob j ec t s " , S ignal Proce s s . : Image
Co mm . Vo l . 6 No . 2 pp . 143 - 161 , 1994 ) や 、 ア ル フ ァ マ ッ プ を 縮小 し て符 号化 し 、 拡大す る 際 に 曲 線近似 す る 方法 (特願平 5 - 2 9 7 1 3 3 号参 照 ) な ど が あ る
上述 の よ う に 画像 を 符 号化す る 場合 に 、 画面 内 を 背 と オブ ジ ェ ク ト に 分割 し て 符号化す る 場 背景 と ォブ ジ ェ ク ト を 分 け る た め に 、 オブ ジ ェ ク ト の 形状や 画面 内 の位置 を表す ア ル フ ァ マ ッ プ信 号が 必要 と な る の た め 、 画像 の 符 号化情報 と 共 に 、 こ の ア ル フ ァ マ ク プの 情報 も 符 号化 し て ビ ッ ト ス ト リ ー ム ( it : ream )化 し 、 伝送や蓄積 に供 さ れ る 。
し か し 、 画面 内 を 背景 と ォブ ジ ェ ク ト に 分割 し て 符 号化す る 方式の場合 、 従来の符 号化法 の よ う に 画面 内 を 一括 し て符 号化す る の に 比べ、 ア ル フ ァ マ ッ プが あ る 分 、 符 号量 ( the number of encoded b i t s )の 増 加 が 問題 と な り 、 こ の ア ル フ ァ マ ッ プの 符 号量 の増加 よ る 符号化効率 の 低下 が 問題 と な る 。 発 明 の 開示
こ の発 明 は 、 オブ ジ ェ ク ト の形状や画 面 内 の位置 な ど を表す副画像情報 で あ る ア ル フ ァ マ ッ プの 情報 を 効 率 良 く 符 号化 で き る と と も に 、 そ の 復 号 を 行 う こ と が で き る よ う に し た 画像符 号化装置お よ び画像復 号化装 置 を提供す る こ と を 目 的 と す る 。
本発 明 に よ る と 、 画像 を そ の 画像 の オブ ジ ェ ク ト 領 域 と 背景領域 に 区別す る た め の 情報 で あ る ア ル フ ァ マ ッ プ と 共 に符 号化 し 、 前記 ア ル フ ァ マ ッ プ は相 対 ァ ド レ ス 符 号ィ匕 ( relative addres s encoding ) を 用 レヽ て 符 号化す る 画像符 号化装置 に お い て 、 既 に符 号化 し た 変化画素 ( change pixel ) を 参照変化画素
( ref erence change pixel ) と し 、 こ の 照 変ィ匕画 素 と 、 次 に符 号化す る 変化画素 の相 対的 な位置 を 表す シ ン ボ ル を 可変長符 号ィヒテ ー ブ ル ( variable- length encoding table )を 用 レヽ て 符 号ィ匕す る 手段 と 、 前記可変長符 号化 テ ー ブ ル を 2 つ 以 上保持 し 、 既 に符 号化 し た 前記 ア ル フ ァ マ ッ プ の ノ、。 タ ー ン に よ っ て 前記可変長符 号化 テ ー ブル を切 り 替 え る 手段 と を備 え る 画像符 号化装置 が提供 さ れ る
本発 明 に よ る と 、 こ の 符 号化装置 に よ り 符号化 さ れ て 得 ら れた符 号化 ビ ッ ト ス ト リ ー ム を 復 号す る 復 号化 装置 で あ っ て 、 前記 シ ン ボ ル を 可変長符 号化 テ ー ブル を 用 い て 復 号す る 手段 と 、 前記可変長符 号化 テ ー ブル を 2 つ 以 上保持 し 、 既 に復 号 し た 前記ア ル フ ァ マ ッ プ の パ タ ー ン に よ っ て 、 前記可変長符 号化 テ ー ブ ル を 切 り 替 え る 手段 と を備 え る 画像符 号化装置 が提供 さ れ る さ ら に は 、 前記可変長符 号化 テ ー ブル を 切 り 替 え る 手段 は 、 参照変化画素 の 近 く の パ タ ー ン に よ っ て 切 り 替 え る 手段 で あ る 。
こ の よ う な構成 の本装置 は 、 変化画素 の位置 を 特定 す る シ ン ボ ル を 可変長符 号化 テ ー ブ ル を 用 い て 符 号化 す る こ と に よ り 符 号量 を 少 な く す る よ う に し た符 号化 z複 号化 に お い て 、 複数種 の 可変長符 号化 テ ー ブル を 用 意 し て お き 、 既 に符 号化 し た 前記 ア ル フ ァ マ ッ プの パ タ ー ン に よ っ て 、 そ の 可変長符 号化 テ ー ブル を 切 り 替 え る こ と を 特徴 と す る も の で あ り 、 こ の よ う な本発 明 に よ れば、 ア ル フ ァ マ ッ プの符 号量 を い っ そ う 低減 で き る 効果が得 ら れ る 。
本発 明 に よ る と 、 時系 列デー タ と し て 得 ら れ る 複数 フ レ ー ム (画像 フ レ ー ム ) の 動画像信 号 を任意形状
( arbitrary shape )の ォブ ジ ェ ク ト 毎 に符 号ィ匕す る 動画像符号化装置 に お け る 、 オブ ジ ェ ク ト の形状 を 表 す ア ル フ ァ マ ッ プ を 符 号化す る 符 号化 回 路 で あ っ て 、 オブ ジ ェ ク ト を含 む方形領域 を M X N 画素 ( M : 水 平 方 向 の 画素数 、 N : 垂直方 向 の 画素数) で構成 さ れ る プ ロ ッ ク 毎 に分割す る 手段 と 、 こ の 分割 さ れて 得 ら れ た 前記 プ ロ ッ ク を 、 前記方形領域 内 に お い て 一定規則 に よ り 順次 、 符 号化す る 手段 と を 有 し 、 ブ ロ ッ ク の 全 て あ る い は一部 に 対 し て 相 対 ア ド レ ス 符 号化 を 適用 す る 2 値画像符号化装置 に お い て 、 プ ロ ッ ク 近傍 の 再生 i ( decoded value ) を 蓄 え る 再生値蓄積手段 と 、 す で に符 号化 さ れた フ レ ー ム (画像 フ レ ー ム ) の 再生信 号 ( decoded s ignal ) を 蓄 え る 画像保持手段 ( フ レ ー ム メ モ リ ) と 、 画像保持手段 ( フ レ ー ム メ モ リ ) 内 の 再生信号 を 用 い て 、 動 き 補償予測値 ( mo t i o n estimation/compens ation alue )を 生成す る 動 き 補償予測 回路 と 、 前記再生値蓄積手段 を 参 照 し 、 ブ 口 ッ ク 近傍の 再生値 も 含 め て 変化画素 を 検 出 す る 手段 と を 有 し 、 相 対ア ド レ ス 符 号化 の 参 照変化画素 を 、 前 記ブ ロ ッ ク 内 の 画素値か ら で な く 、 動 き 補償予測信 号 力 ら 求 め る よ う に し た 2 値画像符号化装置 が 提供 さ れ る 。
ま た 、 M X N 画素 で構成 さ れ る ブ ロ ッ ク 毎 に 、 ォブ ジ ュ ク ト を含 む方形領域 内 を 一定規則 で順次復 号化す る 手段 と 、 ブ ロ ッ ク 近傍 の 再生値 を 蓄 え る 手段 と 、 す で に符 号化 さ れた フ レ ー ム (画像 フ レ ー ム ) の 再生信 号 を 蓄 え る 画像保持手段 ( フ レ ー ム メ モ リ ) と 、 画像 保持手段 ( フ レ ー ム メ モ リ ) 内 の 再生信 号 を 用 い て 、 動 き 補償予測値 を 生成す る 動 き 補償予測 回 路 と 、 プ ロ ッ ク 近傍の 再生値 も 含 め て 変化画素 を検 出 す る 手段 と を 有 し 、 相 対ア ド レ ス 符 号化 の 参照変化画素 を 、 前記 プ ロ ッ ク 内 の 画素値か ら で な く 、 動 き 補償予測信 号 か ら 求 め る よ う に し た ア ル フ ァ マ ッ プ復 号化 回 路 が 提供 さ れ る 。
こ れ に よ り 、 オブ ジ ェ ク ト の 形状や画面 内 の位置 な ど を表す副 画像情報 で あ る ア ル フ ァ マ ッ プの 情報 を 効 率 良 く 符 号化 で き る と と も に 、 そ の 復 号 を行 う こ と 力 S で き る よ う に な る 。
ま た 、 ブ ロ ッ ク 近傍 の 再生値 を 蓄 え る 手段 と 、 す で に符号化 さ れた フ レ ー ム (画像 フ レ ー ム ) の 再生信 号 を 蓄 え る 画像保持手段 ( フ レ ー ム メ モ リ ) と 、 画像 保持手段 ( フ レ ー ム メ モ リ ) 内 の 再生信 号 を 用 い て 、 動 き 補償予測値 を 生成す る 動 き 補償予測 回 路 と 、 プ ロ ッ ク 近傍 の 再生値 も 含 め て 変化画素 を 検 出 す る 手段 と 刖 記フ 口 ッ ク 内 の 再生 ί¾ 素値 ( interpolated ixel or decoded pixel )力 ら 求 め ら れた 、 相 対 ア ド レ ス 符 号化 の 参 照変化画素 と 、 動 き 補償予測信 号か ら 求 め ら れた 、 相 対 ァ ド レ ス 符 号化 の 参 照 変化画素 を 切 り 替 え る 手段 と を 有 し 、 相 対 ア ド レ ス 符 号化情報 を 、 切 り 替 え 情報 と 共 に符 号化す る 画像符 号化装置 が 提供 さ れ る 。
ま た 、 M X N 画素 で構成 さ れ る ブ ロ ッ ク 毎 に 、 ォブ ジ ェ ク ト を含 む方形領域 内 を 一 定規則 で順次複 号化す る 手段 と 、 ブ ロ ッ ク 近傍 の 再生値 を 蓄 え る 手段 と 、 す で に符 号化 さ れた フ レ ー ム ( 画像 フ レ ー ム ) の 再生信 号 を 蓄 え る 画像保持手段 ( フ レ ー ム メ モ リ ) と 、 画像 保持手段 ( フ レ ー ム メ モ リ ) 内 の 再生信 号 を 用 レ、 て 、 動 き 補償予測値 を 生成す る 動 き 補償予測 回 路 と 、 プ ロ ッ ク 近傍の 再生値 も 含 め て 変化画素 を 検 出 す る 手段 と を 有 し 、 前記ブ ロ ッ ク 内 の 再生画素値か ら 求 め ら れた 相 対ア ド レ ス 符 号化 の 参照変化画素 と 、 動 き 補償予測 信 号か ら 求 め ら れた 、 相 対 ア ド レ ス 符 号化 の 参 照変化 画素 を 切 り 替 え る 手段 を 有 し 、 切 り 替 え 情報 に従 っ て 参 照変化画素 を 求 め る よ う に し た ア ル フ ァ マ ッ プ復 号 ィ匕 回 路 が 提供 さ れ る 。
こ の 場合 、 相 対 ア ド レ ス 符 号化 に あ た り 、 参 照 変化 画素 b 1 を 現在処理 中 の 画像 の ブ ロ ッ ク で あ る " current block " 内 よ り 検 出す る 力 、 前 回 処理 し た 画像 の プ ロ ッ ク で あ る " compens ated block" 内 よ り 検 出す る か を 、 プ ロ ッ ク 単位 で切 り 替 え て 処理 す る こ と が で き 、 符 号化側 で は こ の 切 り 替 え 用 の 情報 も 併せて符 号化 し 、 複号化側 で は こ れ を 復号 し て 、 復 号化処理 の 際 に 当 該切 り 替 え 用 の 情報 に 基づ い て 、 参 照変化画素 b 1 を " current block " 内 よ り 検 出 す る 力、 、 " compensated block" 内 よ り 検 出す る 力 を 、 ブ ロ ッ ク 単位 で切 り 替 え る よ う にす る こ と が で き 、 こ の よ う にす る こ と に よ っ て 、 ブ ロ ッ ク 単位 の 画像 内 容 に 基づ い て 、 最適処理が 可能 に な り 、 一層 、 効率 の 良 い符号化 が 可能 に な る 。
ま た本発 明 は 、 時系列デー タ と し て 得 ら れ る 複数 フ レ ー ム の動画像信号 を任意形状 の オブ ジ ェ ク ト 毎 に符 号化す る た め 、 オブ ジ ェ ク ト を含 む方形領域 を M X N 画素 ( M : 水 平方 向 の 画素数 、 N : 垂直方 向 の 画素 数) で構成 さ れ る ブ ロ ッ ク 毎 に 分割す る と 共 に 、 こ の 分割 さ れて得 ら れた前記ブ ロ ッ ク 毎 に 、 前記方形領域 内 を 一 定規則 に よ り 順次 、 符 号化す る 画像符 号化装置 に お い て 、 ォブ ジ ェ ク ト の形状 を 表す ア ル フ ァ マ ッ プ 信号 に つ い て 、 プ ロ ッ ク 近傍の 再生信 号 を 含 む該 フ レ ー ム の 再生信 号 と 、 符 号化 が 終了 し た フ レ ー ム の 再生 信 号 を 蓄 え る フ レ ー ム メ モ リ と 、 プ ロ ッ ク 内 の 画素値 を 2 値の 何れか一方 の値 で全て 置 き 換 え る 手段 と 、 フ レ ー ム メ モ リ 内 のす で に符 号化 が 終 了 し た フ レ ー ム の 再生信 号 を 用 い て 、 動 き 補償予測値 を 生成す る 動 き 補 償予測手段 と 、 プ ロ ッ ク 毎 に 2 値画像 を サ イ ズ変換
(縮小 ' 拡大) す る 手段 と 、 そ の サ イ ズ変換率 ( s i Z e conversion ratio ) の サ イ ド情報 と し て符 号化す る 手段 と 、 ブ ロ ッ ク 毎 に縮小 さ れた 2 値画像 を 符 号化す る 2 値画像符 号化手段 と か ら な る ア ル フ ァ マ ッ プ符 号 ィ匕手段 と を備 え て レ、 る 。
前記 ア ル フ ァ マ ッ プ符 号化手段 は前記ブ ロ ッ ク の 再 生画像 を 、 プ ロ ッ ク 内 を 2 値の何れか一方 の値で全て 置 き 換 え た再生値 と 、 動 き 補償予測値 と 、 ブ ロ ッ ク 毎 に サ イ ズ変換す る 事で得 ら れた 再生値 と の う ち の 何れ か力 ら 選択す る よ う に し た。 従 っ て 、 ア ル フ ァ マ ッ プ 信号の符 号化 を 、 高 品位 で効率 の 良 い形態 で実施 で き . 高 品位 の 画質 を維持 し 、 かつ 、 高圧縮率 で符 号化す る こ と 力 S で き る よ う に な る 。
ま た 、 時系列デー タ と し て 得 ら れ る 複数 フ レ ー ム の 動画像信 号 を任意形状 の ォプ ジ ュ ク ト 毎 に 復 号化す る た め 、 オブ ジ ェ ク ト を含 む M X N 画素 ( M : 水 平方 向 の 画素数 、 N : 垂直方 向 の 画素数) で構成 さ れ る プ ロ ッ ク 毎 に 、 そ の方形領域 内 を 一 定規則 に よ り 順次 、 復 号化す る 画像複号化装置 に お い て 、 ブ ロ ッ ク 近傍 の 再 生信 号 を含む該 フ レ ー ム の 再生信 号 と 、 符 号化済 み フ レ ー ム の 再生信号 を 蓄 え る フ レ ー ム メ モ リ と 、 プ ロ ッ ク 内 の 画素値 を 2 値の 何れか一方 の値で全て 置 き 換 え る 手段 と 、 フ レー ム メ モ リ 内 の す で に符 号化 が 終了 し た フ レ ー ム の 再生信 号 を 用 い て 、 動 き 補償予測値 を 生 成す る 動 き 補償予測 手段 と 、 ブ ロ ッ ク 毎 に 2 値画像 を サ イ ズ変換す る 手段 と 、 ブ ロ ッ ク 毎 に縮小 さ れた 2 値 画像 を 復号化す る 2 値画像復 号化手段 と か ら な る ア ル フ ァ マ ッ プ復 号化手段 を備 え て い る 。
前記ア ル フ ァ マ ッ プ復 号化手段 は 、 前記 ブ ロ ッ ク の 再生画像 を 、 ブ ロ ッ ク 内 を 2 値 の何れか一方 の値 で全 て 置 き 換 え た再生値 と 、 動 き 補償予測値 と 、 ブ ロ ッ ク 毎 に サ イ ズ変換す る こ と で得 ら れた 再生値の う ち の 何 れか か ら 選択す る よ う に し た。 従 っ て 、 高 品 位 の 画像 を 再生 で き る よ う に な る 。
ま た 、 ア ル フ ァ マ ッ プ を ブ ロ ッ ク 毎 に符 号化す る 際 に 、 プ ロ ッ ク 毎 の 属性 ( shape mode ) を 符 号化す る 方 式で あ っ て 、 オブ ジ ェ ク ト を含 む 、 ブ ロ ッ ク サ イ ズの 倍数で表 さ れ る 符 号化領域 (video obj ect plane ( VOP ) )を設定す る 手段 と 、 こ の領域 内 を ブ ロ ッ ク に 分割す る 手段 と 、 各 ブ ロ ッ ク に 対 し て 、 各 々 の 属性 ( shape m o d e )に 固 有 の ラ ベル を 割 り 当 て る ラ ベル付 け 手段 と 、 前記 ラ ベル を フ レ ー ム 毎 に保持す る 記億手 段 と 、 現 フ レ ー ム の 符 号化 ブ ロ ッ ク に 対応す る 前 フ レ ー ム の 参照 ブ ロ ッ ク を 決 定す る 決 定手段 と 、 少 な く と も 前記記憶手段 に保持 さ れた前 フ レ ー ム の ラ ベル と 前 記参 照 プ ロ ッ ク に よ っ て 予測値 を 決定す る 予測 手段 と 前記符 号化 ブ ロ ッ ク の ラ ベル情報 を 、 前記予測値 を 用 い て 符 号化す る 符 号化手段 と を 有 し て 構成す る よ う に し た。
ま た 、 ア ル フ ァ マ ッ プの ブ ロ ッ ク 毎 の 属性 を 再生す る 復 号化装置 と し て 、 復 号化 し た ラ ベル を フ レ ー ム 毎 に保持す る 記憶手段 と 、 現 フ レ ー ム の 復 号プ ロ ッ ク に 対応す る 前 フ レ ー ム の 参照 プ ロ ッ ク を 決定す る 決定手 段 と 、 少 な く と も 前記記億手段 に保持 さ れた 前 フ レ ー ム の ラ ベル と 前記参照 プ ロ ッ ク に よ っ て 予測値 を 決定 す る 予測手段 と 、 前記復 号ブ ロ ッ ク の ラ ベル情報 を 、 前記予測値 を 用 い て 複 号化す る 復 号化手段 を備 え て構 成す る よ う に し た。
こ れ ら 〖こ よ り 、 ァ ノレ フ ァ マ ッ プ を マ ク ロ ブ ロ ッ ク (画像 を 例 え ば、 1 6 X 1 6 画素 と い っ た所定の複数 画素構成 に 分割 し た場合 の そ の 分割 し た 単位画像 プ ロ ッ ク ) 毎 に符号化す る 際 に 、 各 ブ ロ ッ ク の 属性 に 固 有 の ラ ベル を 付 し て こ れ を符 号化す る よ う に し 、 こ の ラ ベル を 再生 し て 元 の ア ル フ ア マ ッ プの デー タ を 再生 ( d e c o d e )す る こ と に よ り 、 効率 の 良 い符 号化 で き る よ う に な る 。
ま た 、 本発 明 は 、 画像 を 、 そ の 画像 の オブ ジ ェ ク ト 領域 と 背景領域 に 区別す る た め の 情報 で あ る ア ル フ ァ マ ッ プ と 共 に符 号化す る に 当 た り 、 ア ル フ ァ マ ッ プ を プ ロ ッ ク 毎 に符号化す る 際 に 、 ブ ロ ッ ク 毎 の 属性 を 符 号化す る よ う に し た方式の 画像符 号化装置 で あ っ て 、 オブ ジ ェ ク ト を含 む 、 ブ ロ ッ ク サ イ ズの倍数 で表 さ れ る 符 号化領域 を設 定す る 手段 と 、 こ の領域 内 を プ ロ ッ ク に 分割す る 手段 と 、 各 プ ロ ッ ク に 対 し て 、 各 々 の 属 性 に 固 有の ラ ベル を割 り 当 て る ラ ベル付 け 手段 と 、 前 記 ラ ベル あ る い は前記 ア ル フ ァ マ ッ プ を フ レ ー ム 毎 に 保持す る 記億手段 と 、 現 フ レ ー ム の符 号化 プ ロ ッ ク に 対応す る 前 フ レ ー ム の 参 照 プ ロ ッ ク を 決 定す る 決 定手 段 と 、 少 な く と も 前記記憶手段 に保持 さ れた 前 フ レ ー ム の ラ ベ ル あ る レヽ は ア ル フ ァ マ ッ プ と 、 前記参 照 プ ロ ッ ク に よ っ て 予測値 を 決定す る 予測 手段 と 、 前記符 号 ィ匕 ブ ロ ッ ク の ラ ベル情報 を 、 前記予測値 を 用 い て 符 号 化す る 符 号化手段 と を 有す る 構成 と す る 。
さ ら に は 、 フ レ ー ム 単位 で の サ イ ズ変換率 を保持す る 記憶手段 を 設 け る と 共 に 、 前記符 号化手段 は 、 フ レ ー ム 単位 で フ レ ー ム の サ イ ズ変換率 が 可変 で あ り 、 前 記サ イ ズ変換率 に 対応 し て 符 号化す る 手段 を備 え 、 ま た 、 前記決定手段 は 、 現 フ レ ー ム の サイ ズ変換率 と 、 刖 記記億手段 か ら 得た前 フ レ ー ム の サ イ ズ変換率 を 用 レ、 て 現 フ レ ー ム の 符号化 プ ロ ッ ク に 対応す る 前 フ レ ー ム の 参照 プ ロ ッ ク を 決定す る 手段 を備 え る よ う にす る 。
あ る い は 、 フ レ ー ム 単位 で の サ イ ズ変換率 を保持す る 記憶手段 を 設 け る と 共 に 、 前記符 号化手段 は 、 フ レ ー ム 単位 で フ レ ー ム の サイ ズ変換率が 可変 で あ り 、 前 記サ イ ズ変換率 に 対応 し て 符 号化す る 手段 を備 え 、 ま た 、 前記決定手段 は 、 現 フ レ ー ム の サイ ズ変換率 と 、 刖 記記億手段 か ら 得た前 フ レ ー ム の サ イ ズ変換率 を 用 レ、 て 現 フ レ ー ム の 符 号化 プ ロ ッ ク に 対応す る 前 フ レ ー ム の 参 照 ブ ロ ッ ク を 決定す る 手段 を備 え 、 ま た 、 前記 予測手段 は 、 参照 ブ ロ ッ ク が複数 あ る 場合 に は 、 前記 複数の 参照 ブ ロ ッ ク の ラ ベ ル の う ち 、 数 が 多 い ラ ベ ル を 予測値 と す る 手段 を備 え る よ う にす る 。
あ る い は 、 フ レ ー ム 単位 で の サ イ ズ変換率 を保持す る 記憶手段 を 設 け る と 共 に 、 前記符 号化 手段 は 、 フ レ ー ム 単位で フ レ ー ム の サ イ ズ変換率 が 可変 で あ り 、 前記サイ ズ変換率 に対応 し て 符 号化す る も の で あ っ て 前 フ レ ー ム あ る い は現 フ レ ー ム あ る い は そ の 両方 の サ ィ ズ変換率 に応 じ て 、 複数種の う ち か ら 選択 し た 一 つ の 可変長符号化テ ー ブル を 用 い て 符 号化 プ ロ ッ ク の符 号化処理 をす る 手段 を 備 え 、 ま た 、 前記決定手段 は 、 現 フ レ ー ム の サ イ ズ変換率 と 、 前記記億手段 か ら 得た 前 フ レ ー ム の サイ ズ変換率 を用 い て 現 フ レ ー ム の 符 号 化 プ ロ ッ ク に対応す る 前 フ レ ー ム の 参 照 プ ロ ッ ク を 決 定す る 手段 を備 え る 。
ま た 、 ア ル フ ァ マ ッ プの ブ ロ ッ ク 毎 の 属性 を 再生す る 複 号化装置 に ぉ レ、 て 、 復 号化 し た ラ ベル あ る い は ァ ル フ ァ マ ッ プ を フ レ ー ム 毎 に保持す る 記憶手段 と 、 現 フ レ ー ム の復 号ブ 口 ッ ク に 対応す る 前 フ レ ー ム の 参 照 プ ロ ッ ク を 決定す る 決 定手段 と 、 少 な く と も 前記記億 手段 に保持 さ れた前 フ レ ー ム の ラ ベル あ る い は ア ル フ ァ マ ッ プ と 前記参 照 プ ロ ッ ク に よ っ て 予測値 を 決定す る 予測手段 と 、 前記復 号ブ ロ ッ ク の ラ ベル情報 を 、 前 記予測値 を 用 い て 復 号化す る 複 号化手段 を 有す る こ と を 特徴 と す る 。
さ ら に は 、 フ レ ー ム 単位 で フ レ ー ム の サ イ ズ変換率 が 可変 で あ り 、 前記サ イ ズ変換率情報 を 複 号化す る 手 段 と 、 前記サイ ズ変換率情報 を保持す る 記憶手段 と を 有 し 、 前記決定手段 は 、 現 フ レ ー ム の サ イ ズ変換率 と 前記記億手段 か ら 読み 出 さ れ る 、 前 フ レ ー ム の サ イ ズ 変換率 を用 い て 現 フ レ ー ム の 復号プ ロ ッ ク に 対応す る 前 フ レ ー ム の 参照 プ ロ ッ ク を 決定す る 機能 を備 え る も の で あ る こ と を 特徴 と す る 。
あ る い は 、 フ レ ー ム 単位 で フ レ ー ム の サ イ ズ変換率 が 可変 で あ り 、 前記サ イ ズ変換率情報 を 複 号化す る 手 段 と 、 前記サイ ズ変換率情報 を保持す る 記憶手段 と を 有 し 、 前記決定手段 は 、 現 フ レ ー ム の サ イ ズ変換率 と 前記記億手段 か ら 読み 出 さ れ る 、 前 フ レ ー ム の サ イ ズ 変換率 を用 い て 現 フ レ ー ム の 復 号ブ ロ ッ ク に 対応す る 前 フ レ ー ム の 参 照 プ ロ ッ ク を 決定す る 機能 を備 え 、 前 記予測手段 は 、 参 照 ブ ロ ッ ク が複数 あ る 場合 に は 、 複 数 の 参照 プ ロ ッ ク の ラ ベルの う ち 、 数 ^ 多 レ、 ラ ベル を 予測値 と す る も の で あ る こ と を 特徴 と す る 。
ま た 、 水平 ' 垂直方 向 共 に 1 / 2 N ( N = 1 , 2 ,
3 , … ) に縮小 さ れた 2 値画像の ブ ロ ッ ク を 拡大す る 拡大 回 路 と し て 、 該 ブ ロ ッ ク 近傍再生値 を保持す る メ モ リ と 、 該ブ ロ ッ ク の縮小率 に応 じ て 前記 メ モ リ に保 持 さ れた再生値 を 1 ノ 2 N に縮小す る こ と で参 照 画 素値 を 求 め る 手段 と 、 水 平 ' 垂直共 に 2 倍 に 拡大す る 処理 を N 回繰 り 返す こ と で 元 の サ イ ズ に 拡大す る 手段 と を備 え 、 前記拡大手段 に お い て は 、 常 に 1 Z 2 N に縮小 さ れた 参 照 画素値 を 用 い る こ と を特徴 と す る 。
ま た 、 時系列デー タ と し て 得 ら れ る 複数 フ レ ー ム の 動画像信 号 を任意形状 の オブ ジ ェ ク ト 毎 に符 号化す る た め 、 オブ ジ ェ ク ト を含 む方形領域 を M X N 画素
( M : 水 平方 向 の 画素数 、 N : 垂直方 向 の 画素数) で 構成 さ れ る プ ロ ッ ク 毎 に 分割す る と 共 に 、 こ の 分割 さ れて 得 ら れた 前記 ブ ロ ッ ク 毎 に 、 前記方形領域 内 を一 定規則 に よ り 順次 、 符 号化す る 画像符 号化装置 で あ つ て 、 前記 フ レ ー ム 内 に お い て 、 オブ ジ ェ ク ト を 含 むブ ロ ッ ク サ イ ズの倍数 で表 さ れ る 領域 を設 定す る 設 定手 段 と 、 こ の 設 定手段 に よ り 設 定 さ れた領域 内 を ブ ロ ッ ク に分割す る 分割 手段 と 、 分割 さ れた前記 プ ロ ッ ク 内 を 動 き 補償予測す る た め に 必要 な動 き べ ク ト ル
( motion vector )を 予 '測符 号ィ匕す る 手段 と を 有す る 画像符 号化装置 に お い て 、 参照 フ レ ー ム 内 の領域 の フ レ ー ム 内 に お け る 位置 を 表す第 1 の位置べ ク ト ル を保 持す る メ モ リ と 、 該参 照 フ レ ー ム 内領域 の フ レ ー ム 内 で の位置 を 表す第 2 の位置べ ク ト ル を 符 号化す る 符 号 化手段 と 、 符号化対象 ブ ロ ッ ク 近傍 の 再生済 み の プ ロ ッ ク の 動 き べ ク ト ル を保持す る 動 き べ ク ト ル メ モ リ と 、 前記動 き べ ク ト ル メ モ リ に 蓄 え ら れて い る 動 き べ ク ト ル を 用 レヽ て 、 符 号化対象 ブ ロ ッ ク の 動 き べ ク ト ル を 予 測す る 手段 と を備 え て な り 、 前記予測 手段 に ぉ レ、 て使用 す る 動 き べ ク ト ルが 、 前 記動 き べ ク ト ル メ モ リ に 存在 し な レヽ場合 に はデ フ オ ル ト の動 き ベ ク ト ル を 予測値 と し 、 こ の デ フ ォ ル ト の 動 き べ ク ト ル は第 1 の位置べ ク ト ル と 第 2 の位置べ ク ト ル と の 差分べ ク ト ル と 、 ゼ ロ べ ク ト ル を 切 り 換 え て 用 い る よ う にす る こ と を 特徴 と す る 。
ま た 、 時系 列デー タ と し て 得 ら れ る 複数 フ レ ー ム の 動画像信号 を任意形状 の オ ブ ジ ェ ク ト 毎 に復 号化 し 、 オブ ジ ェ ク ト を含 む M X N 画素 ( M : 水 平方 向 の 画素 数、 N : 垂直方 向 の 画素数) で構成 さ れ る ブ ロ ッ ク 毎 に 、 そ の 方形領域 内 を 一 定規則 に よ り 順次 、 複 号化す る た め 、 該 フ レ ー ム 内 に お い て 、 オブ ジ ェ ク ト を含 む プ ロ ッ ク サ イ ズの倍数 で表 さ れ る 領域 を プ ロ ッ ク 毎 に 再生す る 画像複 号化装置 に お い て 、 前記 プ ロ ッ ク 内 を 動 き 補償予測す る た め に 必要 な 、 予測符 号化 さ れた動 き べ ク ト ル を 復号す る 手段 と 、 参 照 フ レ ー ム 内 を 動 き 補償予測す る た め に 必要 な 、 予測符 号化 さ れた動 き べ ク ト ル を 復 号す る 手段 と 、 参照 フ レ ー ム 内 の領域 の フ レー ム 内 で の位置 を 表す第 1 の位置べ ク ト ル を保持す る メ モ リ と 、 該 フ レー ム 内 の領域の フ レ ー ム 内 で の位 置 を 表す第 2 の位置べ ク ト ル を 復 号化す る 手段 と 、 復 号化対象 プ ロ ッ ク 近傍 の補正済 みの プ ロ ッ ク の 動 き べ ク ト ル を保持す る 動 き べ ク ト ル メ モ リ と 、 前記動 き べ ク ト ル メ モ リ に保持 さ れた 動 き べ ク ト ル を 用 レヽ て 、 復 号化対象 ブ ロ ッ ク の 動 き べ ク ト ル を 予測す る 予測 手段 と を 有 し 、 前記予測手段 で使用 す る た め の 動 き べ ク ト ルが 、 前記動 き べ ク ト ル メ モ リ に存在 し な い場合 に は デ フ ォ ル ト の 動 き ベ ク ト ル を 予測値 と し 、 こ の デ フ ォ ル ト の動 き べ ク ト ルは第 1 の位置べ ク ト ル と 第 2 の位 置べ ク ト ル と の 差分べ ク ト ル と 、 ゼ ロ べ ク ト ルの いず れカゝ と す る こ と を 特徴 と す る 。 図 面 の簡 単 な説 明
図 1 は 、 本発 明 が 適用 さ れ る 符 号化装置 の概略 プ ロ ッ ク 図 。
図 2 は 、 図 1 の 符 号化装置 に 対応す る 復 号化装置 の 概略プ ロ ッ ク 図。
図 3 は 、 本発 明 が 適用 さ れ る 符 号化装置 の ア ル フ ァ マ ッ プ符号化 回 路 の プ ロ ッ ク 図 。
図 4 は 、 図 3 の 符号化装置 に 対応す る 復 号化装置 に 適用 さ れ る ア ル フ ァ マ ッ プ復 号化 回 路 の構成 を示すブ 口 ッ ク 図。
図 5 は 、 本発 明 の 第 1 の 実施例 に従 っ た符 号化 回 路 の プ ロ ッ ク 図。
図 6 は 、 図 5 の 符 号化 回 路 に 対応す る 複 号化 回 路 の プ ロ ッ ク 図。
図 7 A お よ び 7 B は 、 ブ ロ ッ ク 単位 で符 号化す る 場 合 の 変化画素 の 関係 を 表す図 お よ び b 1 を 検 出す る 為 の 参 照領域 を表す 図 ( プ ロ ッ ク ベー ス 符 号化 の 変化画 素 の 関係 と 参照領域 を表す 図 ) 。 図 8 は 、 M M R を ブ ロ ッ ク ベー ス で符 号化す る 場合 の フ 口 一 チ ヤ一 ト 。
図 9 は 、 図 5 の符 号化 回 路 の 作用 を 説明 す る た め の 図 で あ っ て 、 変化画素 b 1 の 周 囲 の様子 の例 を 示す図( 図 1 0 は 、 図 6 の 復 号回 路 の 作用 を を 説 明 す る た め の 図 で あ っ て 、 参照 画素 の 例 を示す 図 。
図 1 1 は 、 コ ン テ キ ス ト 番 号の 決定方法 を 説 明 す る た め の 図。
図 1 2 は 、 本発 明 の 第 2 実施例 に 従 っ た符 号化 回 路 の ブ ロ ッ ク 図。
図 1 3 は 、 図 1 2 の 符 号化 回 路 に 対応す る 複 号化 回 路' の ブ ロ ッ ク 図。
図 1 4 は 、 本発 明 の 第 3 の 実施例 に か か る 符 号化 回 路 の プ ロ ッ ク 図。
図 1 5 は 、 図 1 4 の符 号化 回 路 に 対応す る 複 号化 回 路 の プ ロ ッ ク 図。
図 1 6 A お よ び 1 6 B は 、 本発 明 の 第 3 実施例 を 説明 す る た め の 図 で あ っ て 、 フ レ ー ム 間符 号化 に お け る 変 ィ匕画素 の検 出 回 路 を説 明 す る 図。
図 1 7 A お よ び 1 7 B は 、 ス キ ャ ン方 向 切 り 替 え を説 明 す る 図。
図 1 8 は 、 本発 明 で用 い る ア ル フ ァ マ ッ プ の 画面 内 を 所定の複数画素構成 に よ る マ ク ロ ブ ロ ッ ク ( M B ) 単 位 に 分割 し た様子 を示す 図 。
図 1 9 は 、 本発 明 の 第 4 の 実施例 に お け る ア ル フ ァ マ ッ プ符 号化 回 路 の ブ 口 ッ ク
図 2 0 は 、 図 1 8 の 符 号化 回 路 に 対応す る ア ル フ ァ マ ッ プ複 号化 回路 の ブ 口 ッ ク 図。
図 2 1 は、 マル コ フ モ デル符 号化 を 説 明 す る た め の 図 図 2 2 A は 、 複数 の 2 値面像符 号化法 を 切 り 換 え て使 用 す る 2 値画像符 号化 回 路 の プ ロ ッ ク 図 。
図 2 2 B は 、 複数 の 2 値面像符 号化法 を 切 り 換 え て 使 用 す る 2 値画像複 号化 回 路 の ブ ロ ッ ク 図。
図 2 3 A お よ び 2 3 B は 、 サ イ ズ変換 (縮小 · 拡大) 処理 に利用 す る 双一次 内 揷 を 説明す る た め の 図。
図 2 4 A お よ び 2 4 B は 、 ス ム ー シ ン グ処理 (平滑化 処理) を説 明す る た め の 図
図 2 5 は 、 本発 明 で用 い る ス ム ー シ ン グ フ イ ノレ タ ( 平 滑化処理 フ ィ ル タ ) の 別 の 例 を 説 明す る た め の 図 。 図 2 6 A お よ び 2 6 B は 、 双 一 次 内 挿 に よ り 水 平 · 垂 直共 に 2 倍 に 拡大す る 処理 の例 を 説明 す る た め の 図 。 図 2 7 A 乃 至 2 7 D は 、 本発 明 で適用 す る 拡大処理 に お け る 内揷画素位置 と 参 照 画素 の利用 範囲 を 説 明 す る 図 2 8 A 乃 至 2 8 B は 、 本発 明 で適用 す る 拡大処理 に お け る 参 照画素 の 追加処理 の 例 を説 明 す る 図 。
図 2 9 は 、 ブ ロ ッ ク 毎 の サ イ ズ変換処理 の 別 の 例 を 説 明 す る た め の 図 。
図 3 0 は 、 プ ロ ッ ク (マ ク ロ プ ロ ッ ク ) を 縦横 " 1 / 2 " の サイ ズ に縮小す る 縮小処理 の一例 を 説 明 す る た め の 図。
図 3 1 は 、 縮小 ブ ロ ッ ク の 画素値 を 求 め る 手法 の例 を 説明 す る た め の 図 。
図 3 2 は 、 画素 間 引 き に よ る 縮小処理 の例 を 説 明 す る た め の 図 。
図 3 3 は 、 本発 明 で用 い る 拡大処理 の 例 を 説 明 す る た め の 図。
図 3 4 A お よ び 3 4 B は 、 本発 明 で使用 す る 双一 次 内 揷処理 に よ り 水 平 · 垂 直共 に 2 倍 に 拡大す る 処理 内 容 を 説 明 す る た め の 図。
図 3 5 は 、 本発 明 の 第 5 の 実施例 に お け る フ レ ー ム 単 位の サ イ ズ変換 と 、 小領域単位 の サ イ ズ変換 を組み合 わせた ア ル フ ァ マ ッ プ符 号化 回 路 の ブ ロ ッ ク 図 。
図 3 6 は 、 図 3 5 の ア ル フ ァ マ ッ プ符 号化 回 路 に 対応 す る ア ル フ ァ マ ッ プ複 号化 回 路 の ブ ロ ッ ク 図。
図 3 7 は 、 フ レ ー ム 毎 の解像度 の例 を 表 し た 図 。
図 3 8 は 、 図 3 5 の符 号化装置 に 必要 な フ レ ー ム メ モ リ を 明示的 に含 め て 表 し た場合 の符 号化装置 の プ ロ ッ ク 図 。
図 3 9 は 、 図 3 6 の複 号化装置 に 必要 な フ レ ー ム メ モ リ を 明 示 的 に含 め て 表 し た場合 の複 号化装置 の プ ロ ッ ク 図。
図 4 0 は 、 図 3 5 の 符 号化装置 に 必要 な フ レ ー ム メ モ リ を 明 示的 に含 め て 表 し た場合 の別 の 符 号化装置 の ブ ロ ッ ク 図 。 図 4 1 は 、 図 3 6 の 復 号化装置 に 必要 な フ レ ー ム メ モ リ を 明 示 的 に含 め て 表 し た場合 の別 の 復 号化装置 の ブ 口 ッ ク 図。
図 4 2 A は 、 本発 明 の符 号化装匱 で用 い る フ レ ー ム メ モ リ の ブ ロ ッ ク 図。
図 4 2 B は 、 本発 明 の復 号化装置 で用 い る フ レ ー ム メ モ リ の プ ロ ッ ク 図 。
図 4 3 A お よ び 4 3 B は 、 本発 明 の 第 6 の 実施例 に 関 連す る 技術の 説 明 を す る た め の 図 。
図 4 4 A お よ び 4 4 B は 、 本発 明 の 第 6 の 実施例 に 関 連す る 技術の説 明 を す る た め の 図 。
図 4 5 A お よ び 4 5 B は 、 本発 明 の 時刻 n — 1 と 時刻 n に お け る フ レー ム 画像 F n - 1 , F n と そ れぞれの フ レ ー ム F n - 1 , F n に お け る 符 号化領域 C A の各 マ ク ロ ブ ロ ッ ク の モ ー ド情報 M D の例 を 説 明 す る た め の 図 。
図 4 6 A お よ び 4 6 B は 、 本発 明 の 第 6 の 実施例 に 関 連す る 技術の 説 明 を す る た め の 図。
図 4 7 A は 、 本発 明 の 第 6 の 実施例 に 関連す る 符 号化 装置 の 説明 を す る た め の 図。
図 4 7 B は 、 本発 明 の 第 6 の 実施例 に 関連す る 復 号化 装置 の 説明 を す る た め の 図。
図 4 8 は 、 本発 明 の 第 6 の 実施例 に お け る 符 号化 回 路 の ブ ロ ッ ク 図 。
図 4 9 A お よ び 4 9 B は 、 時刻 n — 1 と 時刻 n に お け る フ レ ー ム 画像 F n- 1 , F n と そ れぞれの フ レ ー ム F π— 丄 , F n 二 お け る 符 号化領域 C A の 各 マ ク ロ ブ π ッ ク の モ ー ド情報 ( s h a D e mode information) M D の 例 を 説 明す る た め の 図。
図 5 0 A お よ び 5 0 B は 、 符 号化領域 の 変化 と モ ー ド Iff 報 の 対応す る プ ロ ッ ク の位置 の 変化 の様子 を 説 明 す る た め の 図。
5 1 は 、 ア ル フ ァ マ ッ プ を 併用 す る 動画像符 号化装
Jv_ お け る 符 号化デー タ 構成 を 説 明 す る た め の 図 。 図 5 2 A 乃 至 図 5 2 D は 、 符 号化領域 が 占 め る 対象部 分 が フ レ ー ム の 一部分 で あ る 場合 に 生 じ る ラ ベル未定 部分の 対処法 を説 明 す る た め の 図 。
図 5 3 は 、 本発 明 で使用 す る 縮小処理の 例 を 説 明 す る た め の 図。
図 5 4 は 、 本発 明 で使用 す る 縮小処理 を 説 明 す る た め の 図 。
図 5 5 は 、 本発 明 で用 い る ラ ベル予測方法 を 説明 す る た め の 図。
5 6 は 、 本発 明 で使用 す る 縮小 フ レ ー ム か ら 拡大処 に よ り フ レ ー ム を復元す る 処理 の 説 明 をす る た め の 図
図 5 7 は 、 ラ ベル を 予測 に用 い る よ う に し た本発 明 の 号化装置 の プ ロ ッ ク 図。
図 5 8 は 、 本発 明 の 符 号化装置 で用 い る 符 号化処理手 の 例 を 示す フ ロ ー チ ヤ一 ト 。 図 5 9 は 、 本発 明 の 符 号化装置 で用 い る 符 号化処理手 順 の別 の例 を 示す フ ロ ー チ ヤ一 ト 。
図 6 0 は 、 本発 明 の ア ル フ ァ マ ッ プ符 号化 回 路 力ゝ ら 出 力 さ れ る ビ ッ ト 列 の 並び順 を 説明 す る 図。
図 6 1 は 、 本発 明 の 第 6 の 実施例 の シ ス テ ム 構成例 を 示すブ ロ ッ ク 図。
図 6 2 は 、 本発 明 の 第 6 の 実施例 に お け る 処理手順 を 説 明 を す る た め の フ ロ ー チ ヤ一 ト 。
図 6 3 は 、 本発 明 の 第 6 の 実施例 に お け る 処理手順 を 説 明 をす る た め の フ ロ ー チ ャ ー ト 。
図 6 4 は 、 本発 明 に て用 い る 動 き べ ク ト ル の 予測符 号 ィ匕 の 一例 を説明 す る た め の 図。
図 6 5 A お よ び 6 5 B は 、 フ レー ム 内 で の オ ブ ジ ェ ク ト の位置 の動 き が 大 き い場合 で の 、 動 き べ ク ト ル の 予 測 の 不都合 を説明 す る た め の 図 。
図 6 6 A お よ び 6 6 B は 、 本発 明 シ ス テ ム に お け る M V 符 号化 回 路 と そ の 周 辺 回 路 の プ ロ ッ ク 図 。
図 6 7 A は 、 本発 明 シ ス テ ム に お け る M V 再生 回 路 の プ ロ ッ ク 図。
図 6 7 B は 、 本発 明 シ ス テ ム に お け る M V 再生 回 路 の 周 辺 回 路 の ブ ロ ッ ク 図 。 発 明 を 実施す る た め の 最 良 の形態
以 下 、 本発 明 の 実施例 に つ い て 、 図 面 を 参 照 し て 説 明 す る 。 初 め に 、 本発 明 が 適用 さ れ る 画像符 号化 お よ び画像複 号化装置 に つ い て概略 を 説 明 し て お く 。
図 1 は 、 画像 を 符号化す る 場合 に 、 画面 内 を 背景 と ォブ ジ ェ ク ト に 分割 し て 符 号化す る 方式 を 適用 し た 画 像符 号化装置 を 示 し て い る 。 こ の 画像符 号化装置 は 、 差分回 路 ( s u b t r a c t e r ) 1 0 、 動 き 補償予測 回 路
( motion estimation/compensation c ircuit )
( M C ) 1 1 、 直交変換 回 路 ( orthogonal
trans f ormer ) 1 2 、 量子ィ匕 回 路 ( quantizer ) 1 3 、 可変長符 号化 回 路 (variable length coder ) ( V L C ) 1 4 、 逆量子化 回 路 ( dequanti zer ) ( I Q ) 1 5 、 逆直交変換回 路 ( inverse orthogonal
trans f ormer ) 1 6 、 力!]算 回 路 ( adder ) 1 7 、 多重ィ匕 回 路 (multiplexer ) 1 8 お よ びア ル フ ァ マ ッ プ符 号 ィ匕 回 路 ( alpha— map encoder ) 2 0 と 力 ら f 成 さ れ る 。
ア ル フ ァ マ ッ プ符 号化 回 路 2 0 は 、 入力 さ れ る ア ル フ ァ マ ッ プ信号 を 符 号化 し 、 こ の 符 号化 さ れ た 信 号 を ア ル フ ァ マ ッ プ信号 と し て 多重化 回 路
( multiplexer ) 1 8 に 出 力 す る 機能 と 、 こ の ア ル フ ァ マ ッ プ信 号 を復 号 し て 局部復 号信 号 ( local
decoded s ignal ) と し て 出 力 す る 機能 を 有す る 。
特 に 、 本 ア ル フ ァ マ ッ プ符 号化 回 路 2 0 は 、 ブ ロ ッ ク の 単位 で入力 さ れた ア ル フ ァ マ ッ プ信 号 を 符 号化す る に あ た り 、 与 え ら れた縮小率 ( convers ion
ratio ) (倍率) で ア ル フ ァ マ ッ プ の解像度
( resolution )を 下 け る ( down— sc ale )処理 を ft レヽ 、 こ の解像度 を 下 げ る 処理 ( resolution down- scaling proces s ) ;^ な さ れた ァ ゾレ フ ァ マ ッ プ ί言 号 、 即 ち ダ ゥ ンサ ン プノレ さ れた ( down sampled )ア ル フ ァ マ ッ プ信 号 を 符 号化す る と 共 に 、 こ の 符 号化 し た ア ル フ ァ マ ッ プ と 縮小率 の 情報 (倍率情報) と を 多重化 し て こ れを ア ル フ ァ マ ッ プ信 号 と し て 多重化 回 路 1 8 に 出 力 す る 機能 を 有す る 。 局 部復 号信 号 と し て は 、 解像 度 を 下 げ る 処理 、 即 ち ダ ウ ン サ ン プ リ ン グ ( down- s amp 1 i n g )力 S な さ れた ア ル フ ァ マ ッ プ を 元 の解像度 に 戻す処理 を し て 得 た ア ル フ ァ マ ッ プ信 号が用 い ら れ る 。
差分 回 路 1 0 は 、 動 き 補償予測 回路 1 1 よ り 供給 さ れ る 動 き 補償予測信 号 (motion
estimation/compensation s ignal と 人力 [Sj像信 号 ( input video s ignal ) と の 差分 言 号
( di f f erence s ignal ) を 算 出 す る も の で あ り 、 直交 変換回 路 1 2 は 、 差分 回 路 1 0 か ら 供給 さ れた 差分信 号 を 、 ア ル フ ァ マ ッ プ の 情報 に従 っ て 、 直交変換係数 ( orthogonal trans f orm coef f ic ient )に 変換 し て 出 力 す る 。
量子化 回 路 1 3 は こ の 直交変換回 路 1 2 に よ り 得 ら れた 直交変換係数 を 量子化す る 回 路 で あ り 、 可変長符 号化 回路 1 4 は こ の 量子化 回 路 1 3 の 出 力 を符 号化 し て 出 力 す る 回 路 で あ る 。 多重化 回 路 1 8 は こ の 可変長 符 号化 回 路 1 4 に よ り 符 号化 さ れた信 号 と ア ル フ ァ マ ッ プ信 号 と を 、 動 き べ ク ト ル情報等 の サ イ ド情報 と 共 に 多重化 し て ビ ッ ト ス ト リ ー ム と し て 出 力 す る 。
逆量子化 回路 1 5 は 量子化 回 路 1 3 の 出 力 を 逆量子 化す る 機能 を 有 し 、 逆 直交変換回 路 1 6 は こ の 逆 量子 化 回 路 1 5 の 出 力 を ア ル フ ァ マ ッ プ信号 に基 い て 逆直 交変換す る 機能 を 有す る 。 加算 回 路 1 7 は こ の 逆直交 変換回 路 1 6 の 出 力 と 動 き 補償予測 回 路 1 1 カゝ ら 与 え ら れ る 予測信号 (動 き 補償予測信 号) と を加算 し て 動 き 補償予測 回 路 1 1 に 出 力 す る 。
動 き 補償予測 回 路 1 1 は 、 フ レ ー ム メ モ リ を 有 し 、 ア ル フ ァ マ ッ プ符 号化 回 路 2 0 か ら 与 え ら れ る 局部復 号信号 に も と づ い て 動作 し て オブ ジ ェ ク ト 領域 の 信 号、 背景領域の 信号 を 蓄積す る 機能 を 有す る 。 ま た 、 動 き 補償予測 回 路 1 1 は蓄積 し た オブ ジ ェ ク ト 領域 の 画像 カゝ ら 動 き 補償値 を 予測 し て 予測値 と し て 出 力 し 、 ま た 、 蓄積 し た 背景領域 の 画像 か ら 動 き 補償値 ( mo t i o n compensation value ) ¾r 予測 し て 予測値
( prediction value ) と し て 出 力 す る 機能 を 有す る 。
上述 の よ う な構成 の符 号化装置 の 作用 を説 明 す る 。 符号化装置 に は 、 画像信 号 と そ の 画像信号の ア ル フ ァ マ ッ プ信 号が入力 さ れ る 。 こ れ ら 信 号は フ レ ー ム 毎 に そ れぞれ所定画素 サ イ ズ (例 え ば、 M X N 画素 ( M : 水 平方 向 の 画素数 、 N : 垂直方 向 の 画素数) ) の プ ロ ッ ク に 分割 さ れ る 。 こ の 分割処理 は 、 周 知 の 技術 に従 つ て フ レ ー ム 毎 に 画像信 号 を 記憶す る メ モ リ ( 図 示せ ず) を ァ ド レ ス 指 定 し 、 プ ロ ッ ク 単位 で画像信 号 を 読 み 出 す こ と に よ っ て 行わ れ る 。 こ の 分割処理 に よ り 得 ら れた プ 口 ッ ク 単位 の 画像信 号は プ ロ ッ ク 位置順 に信 号線 1 を 介 し て 差分 回 路 1 0 に供給 さ れ る 。 差分 回 路
1 0 で は 、 こ の 入力 (画像信 号) と 、 予測信 号 ( ォ プ ジ ク ト 予測 回路 1 1 か ら の 動 き 補償予測信 号の 出 力 ) と の 差分信号が 算 出 さ れ 、 直交変換回 路 1 2 に 供 給 さ れ る 。
直交変換回 路 1 2 は 、 供給 さ れた 差分信 号 を 、 信 号 線 4 を介 し て ア ル フ ァ マ ッ プ符 号化 回 路 2 0 か ら 供給 さ れ る ア ル フ ァ マ ッ プの 情報 に従 っ て 、 直交変換係数 に 変換 し た後 、 直交変換係数 は量子化 回 路 1 3 に供給 さ れ、 量子化 さ れ る 。 量子化 回 路 1 3 に て 量子化 さ れ て 得 ら れた変換係数 は 、 可変長符 号化 回 路 1 4 に お い て 符 号化 さ れ る と 共 に 、 逆 量子化 回 路 1 5 に供給 さ れ る 。
逆量子化 回 路 1 5 に 供給 さ れた変換係数 は 、 こ こ で 逆量子化 さ れた後 、 逆直交変換回路 1 6 に お い て 逆変 換 さ れ る 。 逆変換係数 は加算 回 路 1 7 に お い て 動 き 補 償予測 回路 1 1 よ り 供給 さ れ る 動 き 補償予測値 と 加 算 さ れ、 局部復 号画像 と し て 出 力 さ れて 、 再び動 き 補償 予測 回路 1 1 に入力 さ れ る 。 加算 回路 1 7 の 出 力 で あ る 局 部復 号画像 は 、 動 き 補償予測 回 路 1 1 内 の フ レ ー ム メ モ リ に蓄 え ら れ る 。
一方 、 こ の 動 き 補償予測 回 路 1 1 は 、 ア ル フ ァ マ ツ プ符 号化 回路 2 0 か ら 与 え ら れ る 局 部復 号信 号 に 基づ レヽ て ォブ ジ ェ ク ト の領域 の ブ ロ ッ ク の 処理 の タ イ ミ ン グで は "オブ ジ ェ ク ト の 動 き 補償予測値 " を 、 ま た 、 そ れ以外の タ イ ミ ン グで は "背景部分 の 動 き 補償予測 値 " を 差分 回 路 1 0 に 出 力 す る 。
す な わ ち 、 動 き 補償予測 回 路 1 1 は 、 ア ル フ ァ マ ツ プ信号の 局 部復 号信 号か ら 現在 、 オブ ジ ェ ク ト の ブ 口 ッ ク 対応部分 の 画像信 号が 差分 回 路 1 0 に入力 さ れて い る の か 、 あ る い は背景部分の プ ロ ッ ク 対応部分 の 画 像信号が 差分 回 路 1 0 に入力 さ れて レ、 る の か を 知 り 、 オブ ジ ェ ク ト の ブ 口 ッ ク 対応部分の 画像信 号 の 入力 期 間 中 で あ れば、 オブ ジ ェ ク ト の 動 き 補償予測信 号 を 、 そ し て 、 背景部分の プ ロ ッ ク 対応部分 の 画像信 号入力 期 間 中 で あれば、 背 景 の 動 き 補償予測信 号 を 、 差分 回 路 1 0 に 与 え る 。
差分 回 路 1 0 で は 、 こ の入力 さ れた 画像信 号 と 、 そ の 画像 の領域対応 の 予測信 号 と の 差 を 算 出す る の で 、 そ の結果、 入力 画像が ォブ ジ ェ ク ト 対応 の領域 の 画像 で あれば、 そ の オブ ジ ェ ク ト の 対応位置 で の 予測値 と 入力 画像 と の 差分信 号が 、 ま た 、 入力 画像 が 背 景 の領 域 の も の で あ れば、 そ の 背景位置対応 の 予測値 と 入力 画像 と の 差分信 号が 算 出 さ れ、 直交変換回 路 1 2 に供 給 さ れ る 。
直交変換回 路 1 2 は 、 供給 さ れた 差分信 号 を 信 号線 4 を介 し て供給 さ れ る ア ル フ ァ マ ッ プの 情報 に従 っ て 、 直交変換係数 に 変換 し た 後 、 量子化 回 路 1 3 に供給す る 。 直交変換係数 は こ の 量子化 回路 1 3 に て 量子化 さ れ る 。
量子化 回 路 1 3 に て 量子化 さ れた変換係数 は 、 可変 長符 号化 回 路 1 4 に お い て 符 号化 さ れ る と 共 に 、 逆量 子化 回 路 1 5 に供給 さ れ る 。 逆量子化 回 路 1 5 に供給 さ れた 変換係数 は こ こ で逆量子化 さ れた 後 、 逆直交変 換回路 1 6 に お い て 逆変換 さ れて加算 回 路 1 7 に供給 さ れ、 動 き 補償予測 回 路 1 1 カゝ ら 加 算 回 路 1 7 に供給 さ れ る 予測値 と 力!] 算 さ れ る 。
加算 回路 1 7 の 出 力 で あ る 局部復 号画像 の 信 号 は 、 動 き 補償予測 回 路 1 1 に供給 さ れ る 。 こ の 動 き 補償予 測 回 路 1 1 は 、 ア ル フ ァ マ ッ プ信 号の 局部復 号信 号か ら 現在 、 カ卩算 回 路 1 7 力ゝ ら ォブ ジ ェ ク ト の ブ ロ ッ ク 対 応 の信号が 出 力 さ れて い る の か 、 あ る い は背景部分 の プ ロ ッ ク 対応 の 信 号が 出 力 さ れて レヽ る の か を 知 り 、 そ の結果 、 ォブ ジ ェ ク ト の ブ ロ ッ ク 対応 の 信 号が 出 力 中 で あ れば、 オブ ジ ェ ク ト 用 の フ レー ム メ モ リ に 、 ま た 、 背景部分 の ブ ロ ッ ク 対応 の 信 号が 出 力 中 で あ れば、 背 景用 の メ モ リ に 与 え る べ く 動作 し て 対応 の メ モ リ に信 号 を 蓄 え る 。 こ れ に よ り 、 ォ ブ ジ ェ ク ト 用 の フ レ ー ム メ モ リ に はオブ ジ ェ ク ト 画像 の みが 、 ま た 、 背景用 の メ モ リ に は背景画像 の み の 画像 が得 ら れ る こ と に な る 。 従 っ て 、 動 き 補償予測 回 路 1 1 は オブ ジ ェ ク ト 画像 を 利用 し て オブ ジ ェ ク ト 画像 の 予測値 を 求 め る こ と が で き 、 ま た 、 背景部分の 画像 を利用 し て 背景画像 の 予測 値 を 求 め る こ と が で き る 。
上述 し た よ う に 、 ア ル フ ァ マ ッ プ符 号化 回 路 2 0 で は 、 入力 さ れ る ア ル フ ァ マ ッ プ を 符 号化 し 、 こ の符 号 化 さ れた ア ル フ ァ マ ッ プ信 号 を信 号線 3 を 介 し て 多重 ィ匕 回 路 1 8 に供給 し て レヽ る 。
ま た 、 多重化 回 路 1 8 に は 、 可変長符 号化 回 路 1 4 か ら 出 力 さ れた 変換係数 が線 4 を 介 し て 供給 さ れて い る 。 多重化 回 路 1 8 は供給 さ れて レ、 る こ れ ら ア ル フ ァ マ ッ プ信号お よ び変換係数 の 符 号化値 は 、 動 き べ ク ト ル情報等 の サ イ ド情報 と 多重化 さ れ た後 、 信 号線 5 を 介 し て 出 力 し て 画像符 号化装置 の 最終出 力 と し て の 符 号ィヒ ビ ッ ト ス ト リ ー ム と な る 。
以 上が符 号化装置 の構成 と 作用 で あ る 。 即 ち 、 こ の 符 号化装置 は 、 こ の 画像 の 誤差信 号 を 得 る に あ た っ て オブ ジ ェ ク ト 用 お よ び背 景用 の 画像 に よ り 動 き 補償予 測 を 行 う べ く 、 ア ル フ ァ マ ッ プ に従 っ て 処理 中 の 画像 の 現在 ブ ロ ッ ク 位置 が オブ ジ ェ ク ト 領域位置 で あ る の か 、 背 景領域位置 で あ る の か を 判 別 し な 力 ら 、 処理 中 の 画像 の 現在 プ ロ ッ ク 位置 が オブ ジ ェ ク ト 領域位置 で あればオブ ジ ェ ク ト 用 の 画像 か ら 求 め た 予測値 を用 い . 背景領域位置 で あ れ ば背景用 の 画像 か ら 求 め た 予測値 を用 い て 差分 を 求 め る 。
オブ ジ ェ ク ト 用 お よ び背景用 の 予測 に は動 き 補償予 測 回 路 に 、 こ の 差分 か ら 得 た 画像 に つ い て 、 ア ル フ ァ マ ッ プ に従 っ て そ れぞれ対応 の領域部分 の 画像 が保持 さ れ、 予測 に供 さ れ る 。 こ れ に よ り 、 オブ ジ ェ ク ト ぉ よ び背景そ れぞれで最適 な 動 き 補償予測 を 行 う こ と が で き 、 質の 良 い 画像圧縮符 号化 と 復 号化 を 可能 にす る 一方 、 図 2 は本発 明 が用 い ら れ る 複 号化装置 の プ ロ ッ ク 図 で あ る 。 復 号化装置 は 、 図 2 に示す よ う に 、 分 離化 回 路 ( demultiplexer ) 3 0 、 可変長複 号化 回 路 ( variable length decoder ) 3 1 、 逆量子ィ匕 回 路 ( dequant i z er ) 3 2 , 逆 直交変換回 路 ( inverse orthogonal trans f orm c ircuit ) 3 3 、 カロ 算 回 路 ( adder ) 3 4 、 動 き 補償 回 路 (motion
compensation c ircuit ) 3 5 N ァ ノレ フ ア マ ッ フ 復 号 ィ匕 回 路 ( alpha - map decoder ) 4 0 と よ り 構成 さ れ る 分離化回 路 3 0 は入力 さ れ る 符 号化 ビ ッ ト ス ト リ ー ム を 分離化処理 ( demultiplex ) し て ア ル フ ァ マ ッ プ 信 号 と 画像 の符 号化信 号等 を 得 る 回 路 で あ り 、 ア ル フ ァ マ ッ プ復 号化 回 路 4 0 は こ の 分離化 回 路 3 0 に て 分 離 さ れた符 号化 ア ル フ ァ マ ッ プ信 号 を復 号 ( decode ) す る 回 路 で あ る 。
可変長復号化 回 路 3 1 は 、 分離化 回 路 3 0 に て 分離 さ れた符 号化画像信号 を 復 号す る 回 路 で あ り 、 逆量子 化 回 路 3 2 は こ の 復 号 さ れた 画像信 号 を 逆量子化 し て 元 の係数 に戻す機能 を 有 し 、 逆直交変換回 路 3 3 は こ の係数 を ア ル フ ァ マ ッ プ に従 っ て 逆 直交変換 し て 予測 誤差信号 に戻す機能 を 有 し 、 加算 回路 3 4 は 、 こ の 予 測誤差信 号 に動 き 補償 回 路 3 5 か ら の動 き 補償値 を加 算 し て 再生画像 言 号 ( decoded video s ignal ) と し て 出 力 す る 。 こ の 再生画像信号が複号化装置 の 最終 出 力 と な る 。
動 き 補償回路 3 5 は 、 加 算 回路 3 4 カゝ ら 出 力 さ れた 再生画像信号 を ア ル フ ァ マ ッ プ に従 っ て フ レ ー ム メ モ リ に 蓄積す る こ と に よ り オブ ジ ェ ク ト 画像 と 背 景画像 と を得 る と 共 に 、 こ の 蓄積 さ れて 得 ら れた画像 か ら ォ ブ ジ ェ ク ト の 動 き 補償信 号 、 背景 の 動 き 補償信 号 を 得 る 。
こ の よ う な構成 の 複 号化装置 に お い て は 、 符 号化 ビ ッ ト ス ト リ ー ム は 、 線 7 を介 し て 分離化 回 路 3 ◦ に供 給 さ れ、 分離化 回 路 3 0 に お い て各 々 の 情報毎 に 分離 さ れ る こ と に よ り 、 ア ル フ ァ マ ッ プ信 号 に 関 す る 符 号 と 、 画像信号の 可変長符 号 と に 分 け ら れ る 。
ア ル フ ァ マ ッ プ信号 に 関す る 符 号は 、 信 号線 8 を 介 し て ア ル フ ァ マ ッ プ復 号化 回 路 4 0 に供給 さ れ 、 ま た 、 画像信 号の 可変長符号 は 可変長複 号化 回路 3 1 に供給 さ れ る 。
ア ル フ ァ マ ッ プ信 号 に 関す る 符 号は ア ル フ ァ マ ッ プ 複 号化 回路 4 0 に お い て ア ル フ ァ マ ッ プ信 号 に 再生 さ れ、 信号線 9 を介 し て 逆 直交変換回 路 3 3 と 動 き 補償 回 路 3 5 に 出 力 さ れ る 。
一方 、 可変長複 号化 回 路 3 1 は 、 分離化 回 路 3 0 か ら 供給 さ れ る 符号 を復 号 し 、 復 号 さ れた 変換係数 は逆 量子化 回 路 3 2 に供給 さ れ、 こ こ で逆量子化 さ れ る 。 逆量子化 さ れた変換係数 は 、 線 9 を 介 し て 供給 さ れ る ア ル フ ァ マ ッ プ に従 っ て 逆直交変換回 路 3 3 に よ り 逆 変換 さ れ、 加算 回 路 3 4 に供給 さ れ る 。 加算 回 路 3 4 は 、 逆直交変換回 路 3 3 力 ら の 逆直交変換 さ れた信 号 と 、 動 き 補償 回路 3 5 よ り 供給 さ れ る 動 き 補償信 号 と を加算 し 、 再生画像 を 得 る 。
以 上が 本発 明 を 適用 す る 画像符 号化装置 お よ び画像 複 号化装置 の概要 で あ る 。
本発 明 は 、 図 1 に示す符 号化装置 の構成要素 で あ る ア ル フ ァ マ ッ プ符 号化 回 路 2 0 お よ び図 2 に示す復 号 化装置 の構成要素 で あ る ア ル フ ァ マ ッ プ復 号化 回 路 4 0 に 関 わ る も の で あ り 、 以 下 に 具体的 な 実施例 を説 明 す る 。
は じ め に 、 変化 画素 の位置 を 特 定す る シ ン ボ ル を 可 変長符 号化テ ー ブル を 用 い て符 号化す る に あ た り 、 出 現頻度 の 多い シ ン ボル に は短 い符 号 を 割 り 当 て る よ う に 可変長符号化テ ー ブル ( V L C テ ー ブル) を 作 っ て く ^ _ と に よ り 、 総符 号量 を 少 な く す る 場合 に 、 既 に 符 号化 /復 号化 さ れた ア ル フ ァ マ ッ プの 画素ノ、。 タ ー ン に よ つ て 、 可変長符 号化テ ー ブル ( V L C テ ー ブル) を 適応的 に切 り 替 え る こ と で 、 さ ら に符 号量 を 削減す る よ う に し た第 1 の 実施例 を 説 明 す る 。
こ の 第 1 の 実施例 は 、 変化画素 の位置 を 特定す る シ ン ボル を 可変長符 号化 テ ー ブル を 用 い て 符 号化 し 、 既 に符 号化 し た ア ル フ ァ マ ッ プの ノ、。 タ ー ン に よ っ て 、 そ の 可変長符 号化テ ー ブル を 切 り 替 え る こ と を 特徴 と し て レヽ る 。
即 ち 、 本実施例 で は 、 既 に符号化 /復号化 さ れた ァ ル フ ァ マ ッ プ の 画素ノ、。 タ ー ン に よ っ て 、 可変長符 号化 テ ー ブル ( V L C テ ー プル) を 適応的 に 切 り 替 え る こ と に よ り 、 さ ら に符号量 を 削減 し て レ、 る 。
先ず 、 図 3 を 参 照 し て 図 1 の 符 号化装置 に用 い ら れ て レ、 る ア ル フ ァ マ ッ プ符 号化 回 路 2 0 の構成 を 説 明 す る 。
ア ル フ ァ マ ッ プ符 号化 回 路 2 0 は 、 解像度 変換回 路 ( fe zjヽ LEI路 ( down-sampling c ircuit ) ) 2 1 、 (拡大 回路 ( up - sampl ing c ircuit ) ) 2 3 、 2 値画 像符 号化 回 路 (binary image encoder ) 2 2 , 多重 ィ匕 回 路 (multiplexer ) 2 4 と 力 ら 構成 さ れて レ、 る 。
こ れ ら の う ち 、 解像度変換 回 路 2 1 は解像度 を 下 げ る ( scal ing-down )た め の 変換回 路 で あ り 、 与 え ら れ る 縮 ,j、拡大率 ( convers ion ratio )に従 つ て ァ ル フ ァ マ ッ プ を縮小 し 、 ま た 、 解像度変換 回 路 2 3 は解像 度 を 上 げ る ( scaling-up )た め の 変換回 路 で あ つ て 、 与 え ら れ る 縮小拡大率 に従 っ て ア ル フ ア マ ッ プ を 拡大 す る 機能 を 有す る 。
解像度変換回 路 2 3 は 、 解像度変換回 路 2 1 が解像 度縮小変換 し た ア ル フ ァ マ ッ プ を 元 の サイ ズ に 戻す た め に設 け て あ り 、 こ の解像度変換回 路 2 3 に よ り 元 の サ イ ズ に戻 さ れた ア ル フ ァ マ ッ プが 、 信 号線 4 を 介 し て 直交変換回 路 1 2 , 逆 直交変換回路 1 6 に 与 え ら れ る ア ル フ ァ マ ッ プ局部復 号信号 と な る 。
2 値画像符号化 回 路 2 2 は 、 解像度変換回 路 2 1 の 出 力す る 解像度縮小変換 さ れた ア ル フ ァ マ ッ プ信 号 ( down-sampled alpha-map s ignal ) を 2 値画像符 号化 し て 出 力 す る 機能 を 有 し 、 多重化 回 路 2 4 は 2 値 画像符 号化 出 力 と 前記信 号線 6 を 介 し て 与 え ら れ る 縮 小拡大率の 情報 と を 多重化 し て 出 力 す る 。
こ の よ う な構成 の ア ル フ ァ マ ッ プ符 号化 回 路 2 0 に お い て は 、 ア ル フ ァ マ ッ プ信号入力線 2 を 介 し て 入力 さ れ る ア ル フ ァ マ ッ プ信 号 を 、 解像度変換回 路 2 1 に よ り 指 定の縮小拡大率 で縮小 し た後 に符 号化 し 、 こ の 符 号化 さ れた ア ル フ ァ マ ッ プ信 号 を信 号線 3 を 介 し て 出 力 し 、 ま た 、 符 号化 さ れた ア ル フ ァ マ ッ プ信号
( down-sampled and encoded alpha-inaD
s ignal )を解像度変換回 路 2 3 に よ り 元 の解像度 に 拡 大 し て 得た局部復 号信号 を 信 号線 4 を介 し て 図 1 の 直 交変換回路 1 2 お よ び逆直交変換回 路 1 6 に 出 力 す る 。
す な わ ち 、 信号線 6 を 介 し て ア ル フ ァ マ ッ プ符 号化 回 路 2 0 に所望 と す る サ イ ズ変換率 ( s i ze
convers ion ratio ) の 設 定情報 を供給す る こ と で 、 前記 ト レ ー ドオ フ ( trade-of f ) を 図 る こ と が 可能 と な る 。
信 号線 6 を介 し て 供給 さ れた サ イ ズ変換率 は 、 解像 度変換回 路 2 1 , 2 3 お よ び 2 値画像符 号化 回 路 2 2 に供給 さ れ、 ア ル フ ァ マ ッ プ信 号の 発 生符 号量 ( t he number of encoded bits )を 制御す る こ と 力 S 可 IS と な る 。 ま た 、 信 号線 6 を 介 し て 供給 さ れた サ イ ズ変 換率 の符 号 i s ize convers ion ration code ) (設 定情報信号) は 、 多重化 回 路 2 4 に て 、 符 号化 さ れた ア ル フ ァ マ ッ プ信号 と 多重化 さ れ、 信号線 3 を 介 し て 出 力 さ れ、 ア ル フ ァ マ ッ プ の 符 号化信 号 と し て 画像符 号化装置 の 最終 出 力 段 で あ る 多重化 回 路 1 8 に 与 え ら れ る こ と に な る 。
次 に 、 図 4 を 参 照 し て 図 2 の 複 号化装置 に用 い ら れ る ア ル フ ァ マ ッ プ復 号化 回 路 4 0 を説 明 す る 。
図 4 に示す よ う に 、 ア ル フ ァ マ ッ プ複 号化 回 路 4 0 は 、 2 値画像複 号化 回 路 4 1 、 解像度変換回 路 4 2 お よ び分離化回 路 4 3 に て 構成 さ れ る 。 分離化 回 路 4 3 は 、 図 2 に示す画像復 号化装置 内 の 分離化 回 路 3 0 で 分離 さ れて ア ル フ ァ マ ッ プ復 号化 回 路 4 0 に 入力 さ れ た ア ル フ ァ マ ッ プ信号カゝ ら ア ル フ ァ マ ッ プ信 号 の 符 号 と サイ ズ変換率の 符 号 ( サ イ ズ変換率 の 設 定情報信 号) に 分離す る ( demultiplex )回 路 で あ り 、 2 値画 像復 号化 回 路 4 1 は ア ル フ ァ マ ッ プ信 号の 符 号 を 、 分 離化 回路 4 3 か ら 分離 し て 与 え ら れ る サ イ ズ変換率 の 符 号 に従 っ て 2 値画像 に 戻す 回 路 で あ り 、 解像度 変換 回 路 4 2 は こ の 2 値画像 を 、 分離化 回 路 4 3 か ら 分離 し て 与 え ら れ る サ イ ズ変換率 の 符 号 に従 っ て 解像度拡 大変換 し て 出 力 す る 。
図 4 に お い て 、 信 号線 8 を介 し て ア ル フ ァ マ ッ プ復 号化 回路 4 0 に供給 さ れた符 号 は 、 分離化 回 路 4 3 に よ り ア ル フ ァ マ ツ プ信 号の 符 号 と サイ ズ変換率 の 符 号 に 分離 さ れ、 各 々 信 号線 4 4 お よ び信 号線 4 5 を介 し て 出 力 さ れ る 。
2 値画像復 号化 回 路 4 1 で は 、 信 号線 4 4 を 介 し て 供給 さ れ る ア ル フ ァ マ ッ プ信 号の 符 号 と 信 号線 4 5 を 介 し て 供給 さ れ る サ イ ズ変換率 の 符 号か ら 、 縮小 さ れ た ァ ノレ フ ァ マ ッ プ信 号 を 再生 し 、 信 号線 4 6 を 介 し て 解像度変換回 路 4 2 に供給す る 。 解像度変換回 路 4 2 で は 、 信 号線 4 5 を 介 し て 供給 さ れ る サ イ ズ変換率 の 符 号か ら 、 縮小 さ れ た ア ル フ ァ マ ッ プ信 号 を 元 の サ イ ズ に拡大 し た後 、 信号線 9 を介 し て 出 力 す る 。
図 5 は 、 図 1 の ア ル フ ァ マ ッ プ符 号化 回 路 2 0 あ る レヽ は 、 図 3 の 2 ィ直画像符号化 回 路 2 2 を よ り 詳細 に表 し て !/、 る 。
ア ル フ ァ マ ッ プ信 号 5 1 は 、 a 1 検 出 回 路 5 2 及び 符 号化済 みの ア ル フ ァ マ ッ プ を保持す る メ モ リ 5 3 に 入力 さ れ る 。 a l 検 出 回 路 5 2 で は 、 図 7 A お よ び図
7 B に示 さ れ る よ う に変化画素 a 1 の位置 が 検 出 さ れ 位置信 号 5 4 を 出 力 す る 。
即 ち 、 図 7 A は 、 ブ ロ ッ ク 単位 (例 え ば、 M X N 画 素構 成 の プ ロ ッ ク 単位 ( M : 水 平方 向 の 画素数 、 N : 垂直方 向 の 画素数 ) ) で ア ル フ ァ マ ッ プ信 号 を 符 号化 す る 場合 の 変化画素 ( changing pixel )の 関係 を 表 す図 で あ る 。 ま た 、 図 7 B は 、 参 照 変化画素
( ref erence changing ixel ) b 1 検 出 す る た め の 参照領域 ( ref erence area )を表す 図 で あ る 。
プ ロ ッ ク ベ ー ス 符 号ィ匕 ( block base encoding ) に お い て は 、 以 下の よ う に 変化画素 は 単純化 し て 符 号化 し て も 良 い。 な お 、 以 下 の 処理 は 、 ス キ ャ ン の順序 を 切 り 換 え て も 良 い し 、 縮小 さ れた プ ロ ッ ク に 適用 し て も 良 い。 単純化 し た 変化 画素 の 符 号化 は 、 次 の よ う に し て行 う 。
今 、 変化画素 a i ( i = 0 〜 1 )お よ び参 照変化 画素 b 1 の そ れぞれ に つ い て の 、 画面左 上カゝ ら ラ ス タ 順 に 求 め た 際 の ア ド レ ス ( ま た は画素順序) を 各 々 、 abs— ai ( i = 0〜 ; L )、 abs— bl と 表記 し 、 変化画素 aO の 属す る ラ イ ン を aO— line と 表記す る と 、 aO— l ine お よ び r— ai ( i = 0〜 ; 1 )、 r— bl のィ直 は 、 以 下 の 式で求 め ら れ る 。
aO— line = ( int ) ( ( abs_a0 + WIDTH ) /
WIDTH ) ― 1
r_a 0 = abs a 0— a 0 l ine * WIDTH
r _a 1 = abs a 1— a 0 line氺 WIDTH
r _b 1 = abs— bl— ( aO— line— 1 ) * WIDTH 前記式 に お い て 、 * は乗 算 を 、 ( i n t ) ( X ) は X の 小数点以 下切 り 捨て を 意味 し て お り 、 WIDTH は プ ロ ッ ク の水 平方 向 の 画素数 を 示 し て い る 。 変化 画素 の 相 対 ア ド レ ス " r— a l— r— bl" あ る レヽ は 、 " r— al— r 一 a 0 " の値 を 符 号化す る こ と で 、 再生値が 得 ら れ る 。 こ の よ う に し て 変ィ匕画素 a 1 の位置 が 検 出 さ れ る 。
上述 の よ う に し て a 1 検 出 回 路 5 2 に よ り 検 出 さ れ た位置 5 4 の 情報 が 、 モ ー ド決定回路 ( shape mode determination c ircuit ) ο 5 に送 ら れ る 。 同 日き に ま た 、 メ モ リ 5 3 か ら は参 照 変化画素 b 1 の位置信 号 5 6 が モ ー ド決定回 路 5 5 に 送 ら れ る 。
モ ー ド決定回 路 5 5 で は 、 図 8 に示す ア ル ゴ リ ズ ム に よ り 、 モ ー ド が 決定 さ れ、 そ の モ ー ド が 、 符 号化 さ れ る シ ン ボ ル 5 7 と し て 符 号化 回路 5 8 に 送 ら れ る 。
即 ち 、 ま ず 、 起点変化画素 の位置 を 初期化 し ( S 1 ) 、 初期位置 ( ブ ロ ッ ク の左 上画素) で の 画素値 を 1 ビ ッ ト で符 号化す る ( S 2 ) 、 次 に初期位置 に お レ、 て 参 照 変化画素 b 1 を 検 出 す る ( S 3 ) 。
こ こ で 、 参照変化画素 b 1 が 検 出 さ れな か つ た場合 に は 、 参照領域 に 変化画素 が 存在 し な い こ と か ら 垂 直 モ ー ド が使 え な い た め 、 垂直ノ、。 ス モ ー ド の 状態 を " T R U E " (真) と し 、 b 1 が 検 出 さ れた場合 は垂 直モ — ド が使 え る た め 、 垂直ノ、 ° ス モ ー ド の 状態 を " F A L
S E " (偽) と す る 。
以 上 で初期状態 の セ ッ テ ィ ン グ を 終了 し 、 符 号化ル ー プの 処理 に移 る 。
ま ず 、 変化画素 a 1 を 検 出 し ( S 5 ) 、 変化画素 a 1 が 検 出 さ れた か否 か を 判 定 し ( S 6 ) 、 変化画素 a 1 が検 出 さ れ な か っ た場合 に は 、 以 後 、 変化画素 が 無 い た め 、 符号化 の 終 了 を 示す符 号化処理終 了 符 号 ( E O M B ; E n d of MB) を 符 号化す る ( S 7 ) 。
ま た 、 ス テ ッ プ S 6 で の判 定の 結果 、 変化画素 a 1 が 検 出 さ れた場合 に は 、 垂直パ ス モ ー ド ( Vertical pass mode) の 状態 を 判 定す る ( S 8 ) 。 こ こ で 、 垂直ノ、。 ス モ ー ド の 状態 が " T R U E " な ら ば、 垂 直パ ス モ ー ド の符号化処理 ( S 1 6 ) を 行い 、 垂 直ノ、。 ス モ ー ド の 状態 が " F A L S E " な ら ば、 b 1 を 検 出 す る ( S 9 ) 。
次 に 、 b 1 が検 出 さ れた か否力 を 判 定 し ( S 1 0 ) . 参 照変化画素 b 1 が 検 出 さ れな か つ た場合 に は 、 水 平 モ ー ド の ス テ ッ プ ( S 1 3 ) に進 み 、 参照 変ィヒ画素 b 1 が検 出 さ れた場合 に は 、 " r — a 1 — r — b 1 " の 絶対値が し き い値 ( V T H ) よ り も 大 き レ、 か否 力 を 判 定 し ( S 1 1 ) 、 そ の 結果 、 し き い値 ( V T H ) 以 下 の場合 に は 、 垂直モ ー ド の ス テ ッ プ ( S 1 2 ) に 進 み、 し き い値 ( V T H ) よ り も 大 き い場合 に は 、 水 平モ ー ド の ス テ ッ プ ( S 1 3 ) に進む。
水平モ ー ド の ス テ ッ プ ( S 1 3 ) で は 、 " r— a 1— r_ a0" の値が符 号化 さ れ る 。 こ こ で 、 " r— al— r— a 0 " の値力 S " W I D T H " よ り も 小 さ い 力 否 力 を 判 定 し ( S 1 4 ) 、 そ の結果 、 " W I D T H " 以 上 の 場 合 に は 、 垂直パ ス モ ー ド の 状態 を " T R U E " (真) と し て ( S 1 5 ) 、 垂直ノ、。 ス モ ー ド の ス テ ッ プ ( S 1 6 ) に 進み 、 垂直ノ ス モ ー ド の ス テ ッ プ ( S 1 6 ) 終 了 し た ら 、 垂直パ ス モ ー ド の 状態 を " F A L S E " (偽) と す る 。
以 上 、 垂直モ ー ド 、 水 平モ ー ド 、 垂直ノヽ。 ス モ ー ド の 何れかが 終了 し た 後 ( a 1 ま で の符 号化 が 終 了 し た 後) 、 a 1 の位置 を新 た な a 0 の位置 と し て ( S 1 8 ) 、 ス テ ッ プ S 5 の 処理 に 戻 る 。
前記の よ う に し て モ ー ド が 決定 さ れ る と 、 メ モ リ 5 3 か ら は 、 符 号化済 み の 参 照変化 画 素 b 1 の 周 囲 の ノ タ ー ン 5 9 が 、 テ ー ブル決定回 路 6 0 に 送 ら れ る 。 テ 一ブル決定回路 6 0 で は 、 複数 の 可変長符 号化テ ー プ ル の う ち の 一つ が 選択 さ れて 出 力 さ れ る 。
こ こ で 、 例 え ば、 図 9 に示す よ う に 、 参 照 変化画素 b 1 の 上方 に ぉ レ、 て 、 右上 力 ら 左 下 向 き の エ ッ ジが あ る 場合 は 、 参照 変化 画素 b 1 の 下方 に も 同 じ エ ッ ジが 直線的 に延び る 場合 が 多 い の で 、 画素 X 1 、 X 2 、 X 3 の 中 で は X 1 に a 1 力 S あ る 確率 力 S 高 レヽ。
そ こ で 、 参照 変化画素 b 1 の 上方が こ の よ う な ノ、。 タ ー ン の 時 に は 、 V L 1 ( r_al-r_bl = -l) に短 い符 号 を割 り 当 て た テ ー ブル を用 い る 。
次 に 、 テ 一 ブル決定方法 を 図 1 0 と 図 1 1 を 参 照 し て 説 明 す る 。 こ こ で は 、 図 1 0 に 示 し た 参 照変化画素 b 1 の 上 2 ラ イ ン の c 0 〜 c 5 に着 目 す る 。 こ れ ら の 画素 が 参照 変化画素 b 1 と 同 じ値 な ら " 1 " 、 異 な る 値 な ら " 0 " と し て 、 図 1 1 に示 し た よ う に . c 0 〜 c 5 の 順序 で " 0 " と " l " を 並べ る 。
こ の 2 進数 を 1 0 進数 に 変換 し た数値 を コ ン テ キ ス ト 番 号 と 呼ぶ。 各 々 の コ ン テ キ ス ト 番 号 に 対応 さ せ て 例 え ば、
[ コ ン テ キ ス ト 番 号 = 0 の 時 ]
V 0 1
V L 1 0 0
V R 1 0
V L 2 0 0 0 0 0
V R 2 0 0 0 0
E O M B 0 0 0 1
H 0 0 1
[ コ ン テ キ ス ト 番 号 の 時 ]
V 0 0 1 0
V L 1
V R 0 0 0 0 1 0
V L 2 0
V R 2 0 0 0 0
E O M B 0 0 0
H 0 0
[ コ ン テ キ ス ト 番 号 = 2 の 時 ]
以 下省略
と い っ た よ う に 可変長符 号化 テ ー ブ ル を 用 意 し て お く こ の テ ー ブル で 、
V L 1 は b を 表 し 、
V L 2 は b - 2 、
V R 1 は b
V R 2 は r — b 1 = 2 を 表す。
図 9 は コ ン テ キ ス ト 番 号 = 1 と な る の で 、 上 の V L 1 1 ビ ッ ト で符 号化 で き る テ ー ブルが 選択 さ れ る 。
再び図 5 に戻 り 、 説 明 を続 け る 。 符 号化 回 路 5 8 は テ一ブル決定回 路 6 0 か ら 送 ら れて く る 選択 さ れ た テ — プル 1 1 を 用 い て符 号 6 2 を 決定 し 、 そ の 決 定 し た 符 号 6 2 を 出 力 す る 。
図 6 は 、 図 2 に示 し た復 号化装置 の構成要素 と し て の ァ レ フ ァ マ ッ プ複 号化 回 路 4 0 あ る レヽ は 、 図 4 に 示 し た復 号化装置 の構成要素 と し て の 2 値画像復 号化 回 路 4 1 を 、 よ り 詳細 に表 し た プ ロ ッ ク 構成 図 を 示す。 こ れは 、 図 5 の 実施例 で生成 さ れ る 符 号 6 2 を 復 号す る 。
符 号 6 2 は復 号化 回 路 6 3 に入力 さ れ る 。 メ モ リ 6 4 に はそ れま で に復 号 さ れた ア ル フ ァ マ ッ プが保持 さ れて お り 、 参 照 変化画素 b 1 の周 囲 の ノ、。 タ ー ン 6 5 が テ ー ブル決定回路 6 6 に送 ら れ る 。
テ一プル決定回 路 6 6 で は 、 複数 の 可変長符 号化 テ 一フ ノレ の う ち の 一 つ が 選択 さ れた テ ー ブル 6 7 と し て 復 号化 回 路 6 3 に 送 ら れ る 。 テ ー ブル決 定の ア ル ゴ リ ズ ム は 図 5 の テ ー ブル決定回 路 7 0 と 同 じ で あ る 。 テ ー ブル 6 7 に よ っ て 、 シ ン ボル 6 8 が復 号 さ れ、 a 1 再生回 路 6 9 に 送 ら れ る 。 a 1 再生 回 路 6 9 で は シ ン ボル 6 8 と メ モ リ 6 4 力、 ら 送 ら れて く る b 1 の位 置 7 0 に よ っ て a 1 の位置 を 求 め 、 a 1 ま で の ァ ノレ フ ァ マ ッ プ 7 1 を 再生す る 。 再生 さ れた ア ル フ ァ マ ッ プ
7 1 は 出 力 さ れ、 ま た 、 今後 の復 号の た め に メ モ リ 6 4 に保持 さ れ る 。
以 上 、 第 1 の 実施例 は 、 複数 の 所定 の 可変長符 号化 テ ー ブル を 切 り 替 え る も の で あ る が 、 図 1 2 に 、 実際 に生起 し た シ ン ボ ル の 頻度 に よ っ て 、 テ ー ブ ル を ダイ ナ ミ ッ ク に修正 し て い く 第 2 の 実施例 を 以 下 に説 明 す る 。
前記第 1 の 実施例 は 、 複数の所定の 可変長符 号化 テ 一ブル を 切 り 替 え る 装置 で あ る が 、 図 1 2 に 、 実際 に 生起 し た シ ン ポ ル の 頻度 に よ っ て 、 テ ー ブル を ダイ ナ ミ ッ ク に修正 し て い く 実施例 が 示 さ れて い る 。 こ れは 図 5 の 第 1 の 実施例 の 符 号化装置 に カ ウ ン タ 7 2 と ノ、 フ マ ンテ ー ブル生成 回 路 ( Hof fman table f orming c ircuit ) 7 3 を カロ え た構成 で あ る 。
カ ウ ン タ 7 2 に はモ ー ド決定回 路 5 5 力ゝ ら の シ ン ポ ル 5 7 と 、 テ ー ブ ル決 定回 路 6 0 力ゝ ら の コ ン テ キ ス ト 番 号 7 4 が 入力 さ れ る 。 カ ウ ン タ 7 2 は 、 各 シ ン ボル の発生回数 を コ ン テ キ ス ト 番 号別 に保持す る 。 こ の保 持カゝ ら 一定の 時 間 が 経過 し た 後 に コ ン テ キ ス ト 番 号別 に各 シ ン ボ ル の発 生 回数 7 5 が ノ、 フ マ ン テ ー ブ ル生成 回 路 7 3 に送 ら れ る 。 ノヽ フ マ 'ン テ一ブル生成 回 路 7 3 で は λ ノヽ フ マ ン符 号化 (藤 田 「基礎情報理論 」 (昭晃 堂) Ρ Ρ . 5 2 - 5 3 、 1 9 8 7 年) に よ っ て 符 号化 テ一ブル 7 6 が 生成 さ れ る 。 そ の テ ー ブル 7 6 が テ ー プル決定回路 6 0 に 送 ら れて 、 該 当 す る コ ン テ キ ス ト 番 号の テ ― ブル力 S テ ー ブル 7 6 で置 き 換 え ら れ る 。 こ のノヽ フ マ ンテ ー プル の 生成 と 置 き 換 え を 全 て の コ ン テ キ ス ト 番号 につ い て 行
図 1 2 の符号化 回 路 で生成 さ れ る 符 号 を 復 号す る た め の 復 号化 回 路 が 図 1 3 に示 さ れて レ、 る 。 図 1 3 に示 す第 2 の 実施例 の 復 号化装置 も 、 や は り 、 図 6 に示す 実施例 の復 号化装置 に 力 ゥ ン タ 7 7 と ノヽ フ マ ンテ 一プ ル生成 回 路 7 8 を加 え た構成 で あ る 。
力 ゥ ン タ 7 7 と ノヽ フ マ ンテ一ブル生成 回 路 7 8 の 動 作 は 2 と 同 じ で あ る 。
以 上述ベた よ う に 、 1 お よ び第 2 の 実施例 は 、 変 化画素 の位置 を 特定す る シ ン ボル を 可変長符 号化 テ ー ブル を用 い て 符 号化す る こ と に よ り 符 号量 を 少 な く す る ぶ う に し た符 号化 /復 号化 に お い て 、 複数種 の 可変 長符 号化テ一ブル を 用 、 し て お き 、 既 に符 号化 し た ァ ル フ ァ マ ッ プの パ タ一ン に よ つ て 、 そ の 可変長符 号化 テ一ブル を 切 り 替 え る こ と を 特徴 と す る も の で あ り 、 こ の よ う な本発 明 に よ れば、 ア ル フ ァ マ ッ プの符 号量 を レヽ つ そ う 低減 で き る 効果 が 得 ら れ る 。
次 に 、 相 対 ァ ド レ ス 符 号化 の 参 照変化画素 を 、 M X N 画素 ( M : 水 平方 向 の 画素数 、 N : 垂 直方 向 の 画素 数) 構成 の ブ ロ ッ ク 内 の 画素値か ら で な く 、 動 き 補償 予測信 号か ら 求 め る よ う に し た実施例 を 第 3 の 実施例 と し て 説明 す る 。
図 1 4 は 、 第 3 の 実施例 と し て の ア ル フ ァ マ ッ プ符 号化 回 路 を 説 明す る プ ロ ッ ク 図 で あ る 。 ま た 、 図 1 5 は 、 第 3 の 実施例 と し て の ア ル フ ァ マ ッ プ復 号化 回 路 を 説明 す る プ ロ ッ ク 図 で あ る 。
図 1 4 、 図 1 5 お よ び図 1 6 A 、 1 6 B を 用 い て 本 発 明 の ア ル フ ァ マ ッ プ符 号化 回 路 2 0 お よ びア ル フ ァ マ ッ プ復号化 回 路 4 0 を 説 明 す る 。
第 3 の 実施例 に お い て は 、 図 1 4 に示す よ う に 、 ァ ル フ ァ マ ッ プ符 号化 回 路 2 0 は 、 解像度変換回 路 (縮 小処理用 回 路 i down-sampling c ircuit ) ) 2 1 、 解像度変換 回 路 (拡大処理用 回 路 ( up- sampling c ircuit ) ) 2 3 、 2 値画像符 号化 回 路 ( binary video encoding c ircuit , f or exam le , block-based MMR encoder ) 2 2 、 多重ィ匕 回 路 ( multiplexer ) 2 4 、 さ ら に動 き 補償予測 回 路
( motion estimation/compensation circuit ) 2 5 お よ び縮小 回 路 ( down— s ampl ing c ircuit ) 2 6 と カゝ ら 構成 さ れて レ、 る 。
こ れ ら の う ち 、 解像度変換回 路 2 1 は解像度縮小変 換 ( down-sampling )用 の 変換回 路 で あ り 、 与 え ら れ る サ イ ズ変換率 ( s ize convers ion ratio )の 設定 情報信 号 に従 つ た縮小率 で ア ル フ ァ マ ッ プ信 号 を符 号 化 し 、 ま た 、 解像度変換回 路 2 3 は解像度拡大変換 ( up- s arapl ing )用 の 変換回 路 で あ っ て 、 与 え ら れ る 拡大率 に従 っ た拡大率 で ア ル フ ア マ ッ プ を 符 号化す る 機能 を 有す る 。
解像度変換回路 2 3 は 、 解像度変換回 路 2 1 が解像 度縮小変換 し た ア ル フ ァ マ ッ プ を 元 の サ イ ズ に 戻す た め に設 け て あ り 、 こ の解像度変換回路 2 3 に よ り 元 の サ イ ズ に 戻 さ れた ア ル フ ァ マ ッ プが 、 信 号線 4 を 介 し て 図 1 の 直交変換回 路 1 2 お よ び逆直交変換回 路 1 6 に入力 さ れ る ア ル フ ァ マ ッ プ局 部復 号信 号 と な る 。
2 値画像符 号化 回 路 2 2 は 、 解像度変換回 路 2 1 が 出 力 す る 解像度縮小変換 ( down-sample ) さ れた ア ル フ ァ マ ッ プ信号 を 2 値画像符 号化 し て 出 力 す る 回 路 で あ っ て 、 詳細 は後述す る が 、 縮小処理用 の解像度変換 回 路 2 6 か ら 信 号線 8 2 に て 供給 さ れ る 解像度縮小変 換 さ れた ア ル フ ァ マ ッ プ の 動 き 補償予測信 号 を利用 し て 符 号化す る 。 ま た 、 多重化 回 路 2 4 は 2 値画像符 号 化 出 力 と 拡大率の 情報 と を 多重化 し て 出 力 す る 。
第 3 の 実施例 に お け る 符 号化 回 路 の構成 は 、 動 き 補 償予測 回 路 2 5 お よ び縮小処理用 の解像度変換回 路 2 6 を備 え て レヽ る 点 が 、 図 3 の 回 路構成 と 異 な り 、 動 き 補償予測 回路 2 5 に は 、 先 に符 号化 さ れた フ レ ー ム の 再生画像 を 蓄積す る フ レ ー ム メ モ リ が設 け ら れ、 拡大 回 路 2 3 よ り 供給 さ れ る 再生信 号 ( decoded s i g n a 1 )力 S 蓄 え る こ と 力 ^ で き る 構成 で あ る と 共 に 、 更 に 、 動 き 補償予測 回 路 2 5 に は 、 動 き べ ク ト ル信 号
( 図 示せず) が 供給 さ れ、 こ の動 き べ ク ト ル信 号 に従 つ て 動 き 補償予測信 号 を 生成 し 、 信号線 8 1 を 介 し て 縮小処理用 の解像度変換回 路 2 6 に供給す る 構成 と し て あ る 。
縮小処理用 の解像度変換回路 2 6 は 、 信号線 8 1 を 介 し て 供給 さ れ る 動 き 補償予測 回 路 2 5 か ら の 動 き 補 償信号 を 、 信 号線 6 を 介 し て 供給 さ れ る サ イ ズ変換率 の 設定情報信号 に応 じ て 縮小 し た 後 、 信 号線 8 2 を 介 し て 2 値画像符号化 回 路 2 2 に 出 力 す る 。
な お 、 2 値画像符 号化 回 路 2 2 と し て 構成す る 場合 で は 、 信号線 2 a を 介 し て供給 さ れ る 解像度変換回 路 2 1 か ら の解像度縮小変換 さ れた ア ル フ ァ マ ッ プ信 号 ( down-s amDled alDha-map s ignal ) を 2 値画像 に 符 号化 し て 出 力 す る 。
こ の よ う な構成 の ア ル フ ァ マ ッ プ符号化 回 路 2 0 に お い て は 、 信 号線 6 を 介 し て 供給 さ れ る サ イ ズ変換率 の 設 定情報信号は 、 解像度変換回 路 2 1 , 2 3 , 2 6 お よ び 2 値画像符 号化 回 路 2 2 に供給 さ れ、 ア ル フ ァ マ ッ プ信 号の発 生符 号量 を 制御す る こ と を 可能 と す る 。 ま た 、 信 号線 6 を 介 し て 供給 さ れた サ イ ズ変換率 の 符 号 (設 定情報信 号) は 、 多重化 回 路 2 4 に て 、 符 号化 さ れた ア ル フ ァ マ ッ プ信 号 と 多重化 さ れ、 信 号線 3 を 介 し て 出 力 さ れ、 ア ル フ ァ マ ッ プ の 符 号化信 号 と し て 画像符 号化装置 の 最終 出 力 段 で あ る 図 1 の 多重化 回 路 1 8 に 与 え ら れ る こ と に な る 。
ア ル フ ァ マ ッ プ符 号化 回 路 2 0 で は 、 解像度変換回 路 2 1 は 、 ア ル フ ァ マ ッ プ信 号入力線 2 を 介 し て 入力 さ れ る ア ル フ ァ マ ッ プ信 号 を 、 信号線 6 を 介 し て 与 え ら れ る 所望 の サ イ ズ変換率 の設定情報 に従 っ て縮小 し ( down sample )、 縮小化 ァ ノレ フ ァ マ ッ プ信 号 ( down- sampled alpha-map s ignal )を 2 直画像符 号ィ匕 回 路 2 2 に 与 え る 。
2 値画像符 号化 回 路 2 2 は 、 解像度変換回 路 2 1 か ら 得 ら れた解像度縮小変換済 みア ル フ ァ マ ッ プ信 号 ( down-sampled alpha-map s i σ n a 1 ) ¾r N 縮小処 ¾ 用 の解像度変換回 路 2 6 か ら 信 号線 8 2 を 介 し て 供給 さ れ る 解像度縮小変換ア ル フ ァ マ ッ プ信 号 の 動 き 補償 予測信 号 を利用 し て 符 号化 し 、 2 値画像符 号化 出 力 と し て 多重化 回路 2 4 と 解像度変換回 路 2 3 と に 与 え る c 多重化 回 路 2 4 は こ の 2 値画像符 号化 出 力 で あ る 符 号 ィ匕 さ れた ア ル フ ァ マ ッ プ信 号 ( encoded alpha-map s ignal ) と 、 信 号線 6 を 介 し て 与 え ら れ る 拡大率 の 情 報 と を 多重化 し て 信 号線 3 に 出 力 す る 。
一方 、 解像度変換回 路 2 3 は 、 こ の 2 値画像符 号化 回路 2 2 か ら 与 え ら れた こ の縮小符 号化 さ れた ア ル フ 了 マ ッ プ信号 ( down-sampled/ encoded alpha-map s ignal ) ( 2 値画像符 号化 出 力 ) を 、 信 号線 6 を 介 し て 得た サ イ ズ変換率 の設 定情報信 号 に従 っ て 元 の 解像 度 の ア ル フ ァ マ ッ プ信 号 に復 号 し 、 そ の 復 号化信 号 を 局 部復 号信 号 と し て 信 号線 4 を 介 し て 動 き 補償予測 回 路 2 5 お よ び図 1 の 直交変換 回 路 1 2 お よ び逆直交変 換回 路 1 6 に 出 力 す る 。
一方 、 動 き 補償予測 回 路 2 5 に は 、 フ レ ー ム メ モ リ が 設 け ら れて お り 、 こ の フ レ ー ム メ モ リ に拡大処理用 の解像度変換 回路 2 3 よ り 供給 さ れ る 先 に符 号化 さ れ た フ レ ー ム の 再生画像信 号 ( revious encoded video f rame s i g n a 1 )力 S 蓄 え ら れ る 。 動 き 補償予測 回 路 2 5 は 、 別途供給 さ れ る 動 き べ ク ト ル信 号 に従 つ て ア ル フ ァ マ ッ プ の 動 き 補償予測信 号 を 生成 し 、 信 号 線 8 1 を 介 し て 縮小処理用 の解像度変換回路 2 6 に供 給す る 。 解像度変換回 路 2 6 は こ の供給 さ れた動 き 補 償予測信 号 を 、 信 号線 6 を 介 し て 得た サ イ ズ変換率 の 設 定情報信号 に従 っ て解像度縮小変換 し 、 2 値画像符 号化 回 路 2 2 に 与 え る 。
2 値画像符 号化 回 路 2 2 は 、 縮小処理用 の 解像度変 換回 路 2 6 か ら 与 え ら れた解像度縮小変換 さ れた ア ル フ ァ マ ッ プ の 動 き 補償予測信 号 を利用 し 、 解像度変換 回 路 2 1 か ら 得 ら れた解像度縮小変換済 みア ル フ ァ マ ッ プ信 号 を符 号化す る 。
以 上が 、 第 3 の 実施例 の ア ル フ ァ マ ッ プ符 号化 回 路 の概要 で あ る 。 次 に 、 ア ル フ ァ マ ッ プ復 号化 回 路 を 以 下 に説 明 す る 。
図 1 5 に 示す よ う に 、 本 実施例 の ア ル フ ァ マ ッ プ復 号化 回 路 4 0 は 、 2 値画像複 号化 回 路 4 1 、 解像度変 換 回 路 (拡大処理用 回 路) 4 2 、 分離化 回 路 4 3 、 さ ら に 動 き 補償予測 回 路 4 4 お よ び解像度変換回 路 (縮 小処理用 回路) 4 5 に て構成 さ れて い る 。
こ れ ら の う ち 、 分離化 回 路 4 3 は 、 図 2 に示す画像 復号化装置 内 の 分離化 回 路 3 0 で分離 さ れ、 ア ル フ ァ マ ッ プ複 号化 回路 4 0 に入力 さ れた ア ル フ ァ マ ッ プ信 号か ら ア ル フ ァ マ ッ プ信 号の 符 号 と サイ ズ変換率 の 符 号 に 分離す る 回 路 で あ り 、 2 値画像復 号化 回 路 4 1 は ア ル フ ァ マ ッ プ信 号の 符 号 を 、 分離回 路 4 3 か ら 分離 し て 与 え ら れ る サ イ ズ変換率 の 符号 (サ イ ズ変換率 の 設 定情報信 号) に従 っ て 2 値画像 に 戻す 回 路 で あ り 、 詳細 は後 述す る が 、 縮小処理用 の解像度変換 回 路 4 5 か ら 信 号線 9 5 に て 供給 さ れ る 解像度縮小変換 さ れた ア ル フ ァ マ ッ プ の 動 き 補償予測信 号 を利用 し て 復 号化 す る 。
拡大処理用 の解像度変換回路 4 2 は こ の 2 値画像復 号化 回 路 4 1 か ら の ア ル フ ァ マ ッ プ信 号の符 号で あ る 2 値画像 を 、 分離化 回 路 4 3 カゝ ら 分離 し て 与 え ら れ る サ イ ズ変換率 の 符号 ( サ イ ズ変換率 の 設 定情報信号) に従 っ て解像度拡大変換 し て 出 力 す る 。
第 3 の 実施例 に お け る 複 号化 回 路 の構成 は 、 動 き 補 償予測 回 路 4 4 お よ び縮小処理用 の解像度変換回 路 4 5 を備 え て い る 点 が 、 図 4 の複 号化 回 路 と 異 な る 。 動 き 補償予測 回 路 4 4 は 、 先 に複 号化 さ れた フ レ ー ム の 再生画像 を 蓄積す る フ レ ー ム メ モ リ を 有 し て お り 、 拡 大処理用 の解像度変換回 路 4 2 よ り 供給 さ れ る 再生信 号 を 蓄 え る と 共 に 、 動 き べ ク ト ル信 号 ( 図 示せず) 力 S 供給 さ れ、 こ の動 き べ ク ト ル信号 に従 っ て 動 き 補償予 測信 号 を 生成 し 、 信号線 9 4 を介 し て 縮小処理用 の解 像度変換回路 4 5 に 供給す る 構成 と な っ て い る 。
解像度変換回 路 4 5 は こ の 動 き 補償予測信 号 を 、 信 号線 9 2 を介 し て 供給 さ れ る サ イ ズ変換率 の 設 定情報 信 号 に応 じ て 縮小 し た 後 、 信 号線 9 5 を 介 し て 2 値画 像複 号化 回路 4 1 に 出 力 す る 。
こ の よ う な構成 の ア ル フ ァ マ ッ プ復 号化 回 路 4 0 に お い て は 、 信号線 8 を介 し て ア ル フ ァ マ ッ プ復 号化 回 路 4 0 に供給 さ れた符 号は 、 分離化 回 路 4 3 に よ り ァ ル フ ァ マ ッ プ信号の 符 号 と サ イ ズ変換率 の 符 号 に 分離 さ れ、 各 々 信 号線 9 1 お よ び信号線 9 2 を 介 し て 出 力 さ れ る 。
2 値画像復 号化 回 路 4 1 で は 、 詳細 は後 述す る が 、 縮小処理用 の解像度変換回 路 4 5 か ら 信 号線 9 5 に て 供給 さ れ る 解像度縮小変換 さ れた ア ル フ ァ マ ッ プ の 動 き 補償予測信号 を利用 し 、 信号線 9 1 を 介 し て 供給 さ れ る ア ル フ ァ マ ッ プ信 号の 符 号 と 信 号線 9 2 を 介 し て 供給 さ れ る サ イ ズ変換率 の 符 号 ( サ イ ズ変換率 の 設 定 情報信 号) に従 っ て 2 値画像 に 戻す復 号化処理 を 施す こ と に よ り 、 縮小 さ れて レヽ る ア ル フ ァ マ ッ プ信 号 を 再 生 し 、 信号線 9 3 を 介 し て 解像度変換回 路 4 2 に 供給 す る 。
解像度変換回 路 4 2 で は 、 2 値画像復 号化 回 路 4 1 の 再生 し た縮小ア ル フ ァ マ ッ プ信 号 を 、 信 号線 9 2 を 介 し て 供給 さ れ る サ イ ズ変換率 の 符 号 に 基づ い て 元 の サ イ ズ に 拡大 し て ア ル フ ァ マ ッ プ信 号 を 再生 し た後 、 信 号線 9 を 介 し て 出 力 す る 。
即 ち 、 解像度変換回 路 4 2 は 、 2 値画像複 号化 回 路 4 1 カゝ ら 与 え ら れた こ の縮小符 号化 さ れた ア ル フ ァ マ ッ プ信 号 ( 2 値画像符 号化 出 力 ) を 、 信 号線 9 2 を 介 し て 得 た サ イ ズ変換率 の 設 定情報信 号 に従 っ て 元 の解 像度 に復 号 し 、 局 部復 号信 号 と し て 得 て 、 こ の得た 局 部復号信 号 を 動 き 補償予測 回 路 4 4 に 出 力 す る 。
一方 、 動 き 補償予測 回 路 4 4 に は 、 フ レ ー ム メ モ リ が設 け ら れ、 拡大処理用 の 解像度変換回 路 4 2 よ り 供 給 さ れ る 、 先 に符 号化 さ れた フ レ ー ム の 再生 画像 を 蓄 積す る 。 動 き 補償予測 回路 4 4 は 、 別途供給 さ れ る 動 き べ ク ト ル信 号 に従 っ て ア ル フ ァ マ ッ プ の 動 き 補償予 測信 号 を 生成 し 、 信 号線 9 4 を 介 し て縮小処理用 の解 像度変換回路 4 5 に供給す る 。 解像度 変換 回 路 4 5 は こ の 供給 さ れた動 き 補償予測信号 を 、 信 号線 9 2 を 介 し て 得た サ イ ズ変換率 の 設 定情報信 号 に 従 つ て 解像度 縮小変換 し 、 2 値画像復 号化 回 路 4 1 に 与 え る 。
2 値画像復 号化 回 路 4 1 は 、 縮小処理用 の解像度変 換回路 4 5 か ら 与 え ら れた解像度縮小変換 さ れた ア ル フ ァ マ ッ プ の 動 き 補償予測信 号 を利 用 し 、 分離化 回 路 4 3 か ら の サ イ ズ変換率 の 設 定情報信 号 に 従 っ て 、 分 離化 回 路 4 3 力ゝ ら の ア ル フ ァ マ ッ プ信 号 を 復 号化す る 以 上が 、 本発 明 を 適用 し た複 号化 回 路 の概要 で あ る 既 に説 明 し て い る よ う に 、 本発 明 を適用 し た第 3 の 実施例 に お け る 符 号化 回 路 の構成 は 、 動 き 補償予測 回 路 2 5 お よ び縮小 回 路 2 6 を備 え て レ、 る 点 が 図 3 の 符 号化 回 路 と 異 な り 、 ま た 、 複 号化 回 路 の構成 は 、 動 き 補償予測 回 路 4 4 お よ び縮小 回 路 4 5 を備 え て い る 点 が 図 4 の 複 号化 回 路 と 異 な る 。
動 き 補償予測 回 路 2 5 ま た は 4 4 に は 、 先 に符 号化 さ れた フ レ ー ム の 再生画像 を 蓄積す る フ レ ー ム メ モ リ が 設 け ら れて お り 、 拡大 回 路 2 3 ま た は 4 2 よ り 供給 さ れ る 再生信号が 蓄 え ら れ る 。 更 に 、 動 き 補償予測 回 路 2 5 ま た は 4 4 に は 、 こ こ に は 図 示 し て レ、 な レ、動 き べ ク ト ル信 号が供給 さ れ、 こ の 動 き べ ク ト ル信 号 に従 つ て 動 き 補償予測信 号 を 生成 し 、 信 号線 8 1 お よ び信 号線 9 4 を介 し て縮小 回 路 2 6 ま た は 4 5 に供給す る こ こ で 、 動 き べ ク ト ル信 号は 、 図 1 お よ び図 2 の装 置 に設 け ら れて い る 、 動 き 補償予測 回 路 1 1 ま た は 3 5 で用 い ら れ る 動 き べ ク ト ル信号 を 禾 IJ用 し て も 良 レヽ し ア ル フ ァ マ ッ プ符 号ィヒ 回 路 2 0 に お い て 、 ア ル フ ァ マ ッ プ用 の 動 き べ ク ト ル検 出 回 路 を 設 け ら れ る こ と に よ り 、 ア ル フ ァ マ ッ プ用 の動 き べ ク ト ル信号 を 求 め て も 良 い。
す な わ ち 、 動 き 補償予測 回 路 2 5 ま た は 4 4 に 供給 さ れ る 動 き べ ク ト ル信 号の 求 め 方 は種 々 知 ら れて お り 、 本発 明 に 関 わ る も の で は な い た め 、 こ こ で は こ れ以 上 言及 し な い。
縮小 回 路 2 6 ま た は 4 5 に お い て は 、 信 号線 8 1 お よ び信 号線 9 4 を 介 し て 供給 さ れ る 動 き 補償信 号 を 、 信 号線 6 お よ び信 号線 9 2 を 介 し て供給 さ れ る サ イ ズ 変換率 の 設 定情報信 号 に応 じ て 縮小 し た 後 、 信 号線 4 2 お よ び信 号線 9 2 を 介 し て 出 力 す る 。
な お 、 2 値画像符 号化 回 路 2 2 と し て構成す る 場合 で は 、 信 号線 2 a を 介 し て 供給 さ れ る 解像度縮小変換 さ れた ア ル フ ァ マ ッ プ信 号 を 2 値画像符 号化 し て 出 力 す る 。
こ こ で 、 本実施例 に か か る 2 値画像符 号化 回 路 2 2 が 図 3 の符 号化 回 路 と 根本的 に 異 な る 点 は 、 信 号線 8 2 を 介 し て 供給 さ れ る 、 解像度縮小変換 さ れた ア ル フ ァ マ ッ プの動 き 補償予測信 号 を利用 し て 符 号化す る 機 能 を 有す る 。
こ の こ と に つ レヽ て 、 詳 し く 説 明 す る 。
図 1 6 A お よ び 1 6 B は 、 動 き 補償予測信 号 を利用 し て符 号化す る 方法 を 説 明 す る 図 で あ り 、 フ レ ー ム 画 像単位 の 画像 に お け る 分割 さ れた N X M画素構成 の 画 像 プ ロ ッ ク の う ち の 一つ を 示 し て レ、 る 。
図 1 6 A お よ び 1 6 B に お レ、 て 、 " c u r r e n t
b l o c k " は処理対象 ブ ロ ッ ク で あ り 、 入力 さ れた 現在 の 処理画像の ブ ロ ッ ク で あ る 。 ま た 、 " compensated block" はネ甫償 ブ ロ ッ ク で あ り 、 前 回処理対象 と な っ た 画像 の プ ロ ッ ク で あ る 。
第 1 の 実施例 で は 、 現在 の 処理対象画像の ブ ロ ッ ク に対応す る ア ル フ ァ マ ッ プ の該 当 ブ ロ ッ ク 上 で の 参 照 変化画素 b l を 変化画素 a 0 、 a 1 と 同 じ " current block" 内 で検 出 し て い た。
一方 、 図 1 4 お よ び 1 5 の 第 3 の 実施例 で は 、 参 照 変化画素 b 1 は動 き 補償予測信号で あ る
" compensated block" 内 力、 ら 検 出 さ れて お り 、 こ の 点 が新 し い概念 で あ る 。 す な わ ち 、 現在 の 処理対象 画像の ブ ロ ッ ク に 対応す る ア ル フ ァ マ ッ プの 該 当 ブ ロ ッ ク 上 で の 参照変化画素 b 1 を 、 動 き 補償予測信 号で あ る compensated block 内 力 ら 検 出 し て レヽ る 。
な お 、 本実施例 は 、 参 照 変化画素 b 1 の検 出 手段 が 異 な る だ け で 、 a 0 、 a 1 お よ び b 1 の相 対 ア ド レ ス を用 い て 符 号化 · 復 号化 を 行 う 点 は 、 先 の 実施例 と 同 一 で あ る 。
図 1 6 A お よ び 1 6 B に お い て 、 a 0 は起点変 化画 素 で あ り 、 すで に 起点変化画素 a 0 ま で は符 号化 が 済 ん で い る 。 ま た 、 a 1 は 、 起点変化画素 a 0 の 次 の 変 ィ匕画素 で あ り 、 bO は 、 " compensated block" 内 に お レ、 て 、 a 0 と 同 じ位 置 の 画素 ( 変化 画素 と は 限 ら な レ、 ) で あ る 。 ま た 、 aO ( bO )力 属す る ラ イ ン を " aO - line" と 表記す る と 、 参 照変化画素 b 1 は次 の よ う に 定義 さ れ る 。 こ こ で 、 a b s— X を プ ロ ッ ク 左 上 の 画素 カゝ ら ブ ロ ッ ク 内 を ラ ス タ 順 に ス キ ャ ン し た 際 の 、 画素 X の ァ ド レ ス と す る 。 な お 、 ブ ロ ッ ク 左 上の 画素 の ア ド レ ス は
" 0 " と す る 。
abs b 0 < abs_b 1 で あ り 、 符 号 " X " を 付 し て 示 す画素 は変化画素 で あ っ て 、 変化画素 X が " a 0 - line" 上 に あ る 場合 に は 、 " a 0 " と 反 対色 の 最初 の 変加画素 を 参 照変化画素 b 1 と し 、 変化画素 力 " a 0 - line" 上 に な い 場合 は 、 該 ラ イ ン 上 の 最初 の 変カロ 画素 を 参照 変化画素 b 1 と す る 。
図 1 6 A は変化画素 が " a 0— 1 i n e " 上 に な い場合 で あ り 、 こ の場合 、 次 の ラ イ ン の最初 の 変化画素 を
" bl " と し て レヽ る 。
ま た 、 図 1 6 B は 、 変化画素 が " a 0— 1 i n e " 上 に あ る 場合 で あ る が 、 こ の 変化画素 X は 、 " a " と 反対色 で は な い の で " b 1 " と はせず に 、 次 の ラ イ ン の 最初 の 変化画素 を " b 1 " と し て レ、 る 。
な お 、 " a0 - line "お よ び " r a i ( i = 0 〜 1 ) "、 " r_b 1 "の値は 、 以 下 の 式で求 め ら れ る 。
a 0 - line = ( int ) ( ( abs - a 0
+W I DTH ) /WIDTH ) - 1
r - a 0 = abs - a 0 - a 0 - line*WIDTH
r - a 1 = abs - a 1 - a 0 一 line*WIDTH
r - bl = abs - b 1 - a 0 - 1 ine *WI DTH
前記式 に お い て 、 * は乗 算 を 、 ま た 、 ( i n t ) ( X ) は X の 小数点以 下切 り 捨て を 意味 し て お り 、 WIDTH は ブ ロ ッ ク の 水 平方 向 の 画素数 を示 し て い る 。
本発 明 で は 、 参 照変化画素 b 1 の 定義 が先 の 実施例 と は変 わ っ た た め 、 " r— bl" の 定義 も 、 上式の よ う に 変 更 さ れ る 。
図 1 6 A お よ び 1 6 B を 用 い て 説 明 し た符 号化方法 は 、 " compensated block" 内 力、 ら 参 照変ィ匕画素 b 1 を 求 め る 方法 の 一例 で あ り 、 参 照 変化画素 b 1 の 検 出 に っ レ、 て は種 々 変形 が 可能 で あ る 。
ま た 、 2 値画像複 号化 回 路 4 1 は 、 信 号線 9 5 を 介 し て供給 さ れ る 解像度縮小変換 さ れた ア ル フ ァ マ ッ プ の 動 き 補償予測信 号 ( " compens ated block" ) を 利用 し て 、 2 値画像符 号化 回 路 2 2 と 同 一 の 手順 で参 照 変化画素 b 1 を 検 出 で き る 。
更 に 、 参 照 変化画素 b 1 を " current block " 内 よ り 検 出 す る 力、、 " compensated block" 内 よ り 検 出す る 力、 を 、 た と え ばプ ロ ッ ク 単位 で切 り 替 え る よ う にす る こ と も で き る 。 こ の 際 、 2 値画像符号化 回 路 2 2 で は切 り 替 え用 の 情報 も 併せ て 符 号化 し 、 2 値画像 復 号化 回 路 4 1 は切 り 替 え用 の 情報 も 復 号化 し て 、 復 号化処理の 際 に は切 り 替 え 情報 に 基づ い て 、 参 照 変化 画素 1 を " current block" 内 よ り 検 出 す る 力 、 " compensated block" 内 よ り 検 出 す る 力 を 、 た と え ばブ ロ ッ ク 単位 で切 り 替 え る よ う にす る 。
こ の よ う にす る と 、 ブ ロ ッ ク 単位 の 画像 内 容 に 基づ レ、 て 、 最適処理 が 可能 に な り 、 一層 、 効率 の 良 い符 号 化 が 可能 に な る 。
ま た 、 ス キ ャ ン順序 を 切 り 替 え る 手段 を 設 け て 図 1 7 A に示す よ う に 、 ス キ ャ ン順序 を横方 向 ス キ ャ ン に 切 り 替 え た り 、 図 1 7 B に示す よ う に 、 縦方 向 ス キ ヤ ン に切 り 替 え る よ う にす る こ と で 、 変化画素 の 数 が 減 り 、 更 に符 号量が 削 減 さ れて 、 こ れ も 一層 、 効率 の 良 い符 号化 につ な が る 。
以 上 、 第 3 の 実施例 は 、 時系列デー タ と し て 得 ら れ る 複数 フ レ ー ム の 動画像信 号 を任意形状 の オ ブ ジ ェ ク ト 毎 に符号化す る 動画像符 号化装置 に お け る 、 ォ ブ ジ ェ ク ト の形状 を 表す ア ル フ ァ マ ッ プ を符 号化す る 符 号 化 回 路 を提供す る 。
即 ち 、 オブ ジ ェ ク ト を含 む方形領域 を M X N 画素 ( M : 水 平方 向 の 画素数 、 N : 垂 直方 向 の 画素数) の ブ ロ ッ ク の 単位で分割す る こ と に よ っ て 得 ら れ る 複数 の プ ロ ッ ク を 、 方形領域 内 に お い て 一 定規則 に よ り 順 次 、 符 号化す る と 共 に プ ロ ッ ク の 全て あ る い は一部 に 対 し て 相 対ァ ド レ ス 符 号化 を 行 う 符 号化器 と 、 ブ 口 ッ ク 近傍の 再生値 を 蓄 え る メ モ リ と 、 す で に符 号化 さ れ た フ レ ー ム の 再生信 号 を 蓄 え る フ レ ー ム メ モ リ と 、 フ レ ー ム メ モ リ 内 の 再生信 号 を 用 い て 動 き 補償予測値 を 生成す る 動 き 補償予測 回路 と 、 ブ ロ ッ ク 近傍 の 再生値 も 含 め て 変化画素 を 検 出す る 検 出 回 路 と を 有 し 、 相 対 ア ド レ ス 符 号化 の 参 照 変化画素 を 、 ブ ロ ッ ク 内 の 画素 値 か ら で な く 、 動 き 補償予測信 号か ら 求 め ら れ る が 動 画像符 号化装置 が 提供 さ れ る 。
ま た 、 ア ル フ ァ マ ッ プ復 号化 回 路 に お い て 、 M X N 画素 で構成 さ れ る ブ ロ ッ ク 毎 に 、 オブ ジ ェ ク ト を 含 む 方形領域 内 を 一 定規則 で順次復 号化す る 複 号化 回 路 と 、 プ ロ ッ ク 近傍 の 再生値 を 蓄 え る メ モ リ と 、 す で に符 号 化 さ れた フ レ ー ム の 再生信 号 を 蓄 え る フ レ ー ム メ モ リ と 、 フ レ ー ム メ モ リ 内 の 再生信号 を 用 い て 、 動 き 補償 予測値 を 生成す る 動 き 補償予測 回 路 と 、 ブ ロ ッ ク 近傍 の 再生値 も 含 め て 変化画素 を検 出す る 検 出 回 路 と を 有 し 、 相 対 ア ド レ ス 符 号化 の 参 照 変化画素 を 、 前記 プ ロ ッ ク 内 の 画素値か ら で な く 、 動 き 補償予測信 号か ら 求 め る 動画像符 号化装置 が提供 さ れ る 。
こ れ に よ り 、 オブ ジ ェ ク ト の 形状や画面 内 の位置 な ど を表す副 画像情報 で あ る ア ル フ ァ マ ッ プの 情報 を 効 率 良 く 符 号化 で き る と と も に 、 そ の 復 号 を 行 う こ と 力 S で き る よ う に な る 。
ま た 、 ブ ロ ッ ク 近傍の 再生値 を 蓄 え る 再生値蓄積回 路 と 、 すで に符号化 さ れた フ レ ー ム の 再生信 号 を 蓄 え る フ レ ー ム メ モ リ と 、 フ レ ー ム メ モ リ 内 の 再生信 号 を 用 い て動 き 補償予測値 を 生成す る 動 き 補償予測 回 路 と 、 再生値蓄積回 路 の 蓄積再生値 の 情報 を 参 照 し 、 プ ロ ッ ク 近傍の 再生値 も 含 め て 変化画素 を検 出 す る 検 出 回 路 と 、 ブ ロ ッ ク 内 の 再生画素値か ら 求 め ら れた 、 相 対 ァ ド レ ス 符 号化 の 参 照変化画素 と 、 動 き 補償予測信 号か ら 求 め ら れた 、 相 対 ア ド レ ス 符 号化 の 参 照 変化画素 を 切 り 替 え る 切替 回 路 と を 有 し 、 相 対 ア ド レ ス 符 号化情 報 を 、 切 り 替 え 情報 と 共 に符 号化す る 動 画像符 号化装 置 が 提供 さ れ る 。
ま た 、 ァ ル フ ァ マ ッ プ復 号ィ匕 回路 に お い て 、 M X N 画素 で構成 さ れ る ブ ロ ッ ク 毎 に 、 オ ブ ジ ェ ク ト を含 む 方形領域 内 を 一 定規則 で順次復 号化す る 符 号化器 と 、 ブ 口 ッ ク 近傍の 再生値 を 蓄 え る 再生値蓄積 回 路 と 、 す で に符 号化 さ れた フ レ ー ム の 再生信 号 を 蓄 え る フ レ ー ム メ モ y と 、 フ レ ー ム メ モ リ 内 の 再生信 号 を 用 い て 動 き 補償予測値 を 生成す る 動 き 補償予測 回 路 と 、 ブ ロ ッ ク 近傍の 再生値 も 含 めて 変化画素 を 検 出 す る 検 出 回 路 と を 有 し 、 ブ 口 ッ ク 内 の 再生画素ィ直カゝ ら 求 め ら れた 、 相 対 ァ ド レ ス 符 号化 の 参 照 変化画素 と 、 動 き 補償予測 信 号か ら 求 め ら れた 、 相 対ア ド レ ス 符 号化 の 参 照 変化 画素 を切 り 替 え る 切替 回 路 を 有 し 、 切 り 替 え 情報 に従 つ て 参 照 変化画素 を 求 め る よ う に し た動画像符 号化装 置 が 提供 さ れ る 。
こ の お 、 相 対 ア ド レ ス 符 号化 に あ た り 、 参 照 変化 画素 b 1 を " current block" 内 よ り 検 出 す る 力 、 compensal: ed block" 内 よ り 検 出 す る 力 を 、 ブ ロ ッ ク 単位 で切 り 替 え て 処理す る こ と が で き 、 符 号化側 で は こ の切 り 替 え用 の 情報 も 併せ て 符 号化 し 、 復 号化 側 で は こ れ を復 号化 し て 、 符 号化 さ れた ア ル フ ァ マ ツ プの 復 号化処理 の 際 に 当 該切 り 替 え 用 の 情報 に 基づ い て 、 参 照変化画素 b 1 を " current block" 内 よ り 検 出 す る か 、 " compensated block" 内 よ り 検 出 す る 力 を 、 ブ ロ ッ ク 単位で切 り 替 え る よ う にす る こ と 力 S で き 、 こ の よ う にす る こ と に よ っ て 、 ブ ロ ッ ク 単位 の 画像 内 容 に基づい て 、 最適処理 が 可能 に な り 、 一層 、 効率 の 良 い符 号化 が 可能 に な る 。
以 上 、 本発 明 に よ れば、 オブ ジ ェ ク ト の 形状や画 面 内 の位置 な ど を表す副 画像情報 で あ る ア ル フ ァ マ ッ プ の 情報 を 効率 良 く 符 号化 で き る と と も に 、 そ の 復 号 を 行 う こ と が で き る よ う に し た 画像符 号化装置 お よ び画 像複号化装置 が得 ら れ る 。
以 上の 実施例 は 、 ア ル フ ァ マ ッ プ符 号化 回 路 2 0 は 、 M M R ( Modif ied Modi f ied READ) 符 号ィ匕 を 用 い た場合の例 を 説明 し た。 し か し 、 本発 明 は M M R 符 号 化 に 限定 さ れ る も の で は な く 、 他 の任意 の 2 値画像符 号化 回路 を 用 い て も 実現可能 で あ る 。 そ こ で 、 そ の よ う な 例 を 以 下説明 す る 。
図 1 8 , 図 1 9 お よ び図 2 0 を 参 照 し て ア ル フ ァ マ ッ プ符 号化 回 路 2 0 お よ びア ル フ ァ マ ッ プ復 号化 回 路 4 0 の 具体的 な構成 を説 明 す る 。
図 1 8 は 、 ア ル フ ァ マ ッ プの 画面 内 を例 え ば、 1 6 X I 6 画素 と い っ た所定の複数画素構成 に よ る マ ク ロ プ ロ ッ ク ( M B ) 単位 に 分割 し た 図 で あ り 、 正方形 の 升 目 の 枠 が 分割 の境界線で あ っ て 、 升 目 一つ 一つ が マ ク ロ ブ ロ ッ ク ( M B ) で あ る 。 2 値で表現 さ れた ア ル フ ァ マ ッ プの 場合 ( オブ ジ ェ ク ト を合成す る 際 の 重 み係数 も 含 め て 多値で表 現 さ れ る 場合 も あ る ) 、 オブ ジ ェ ク ト の 形状情報 は画素毎 に 透明 ( transparent )を 表すィ直力 不透明 ( opaque )を表 す値の いずれかで表 さ れ る 。 従 っ て 、 図 1 8 に示 さ れ る よ う に 、 ア ル フ ァ マ ッ プの 画面 に お け る 各 マ ク ロ ブ ロ ッ ク ( M B ) の 中 身 の 状態 は 、 " transparent"
( 全 て ; ι≤| 明 every pixel in the MB is
transparent ) ) 、 o aque" (全 て 小透明 ( every pixel in t e MB is o aque ) ) 、 Mu 11 i 、 て の他) の 3 種類 の いずれか に 分類 さ れ る 。
人物像の ア ル フ ァ マ ッ プで あ る 図 1 8 の よ う な 画 面 の場合 、 背景 は "透 明 " 、 人物部分 は "不透 明 " で あ る 力ゝ ら 、 " Mu 11 i " に 分類 さ れ る オブ ジ ェ ク ト の 境界 部分 力 S含 ま れ る マ ク ロ ブ ロ ッ ク ( M B ) の み 、 2 値画 像符 号化すれば よ い。 ま た 、 " Mu 1 t i " に 分類 さ れ る ブ ロ ッ ク ( M B ) の 中 で も 、 動 き 補償予測誤差値が 設 定値 ( し き い値) 以 下 の場合 に は 、 該 ブ ロ ッ ク ( M B ) を 動 き 補償予測値で コ ピーす る 。 こ の よ う に 、 コ ピ ー さ れ る マ ク ロ ブ ロ ッ ク ( M B ) の モ ー ド を " no update" 、 2 値面像符 号化 さ れ る マ ク ロ ブ ロ ッ ク
( M B ) の モ ー ド を " coded" と 表記す る と 、 該マ ク ロ ブ ロ ッ ク ( M B ) の 符 号ィヒモ ー ド は次 の 4 通 り と な る 。
( 1 ノ transparent" ( 2 ) o aque
( 3 ) no update
( 4 ) c o d e α
そ れぞれ の モ ー ド の 符 号化法 あ る い は複 号化 法 に つ い て は 、 ア ル フ ァ マ ッ プ符 号化 回 路 2 0 、 ア ル フ ァ マ ッ プ復 号化 回 路 4 0 の 中 で説明 す る 。
図 1 9 は 、 ア ル フ ァ マ ツ プ符 号化 回 路 2 0 の 詳細 な 構成 図 で あ る 。 図 1 9 の構成 に お い て は 、 モ ー ド判 定 回 路 1 1 0 、 C R (サ イ ズ変換率) 判 定回 路 1 1 1 、 セ レ ク タ 1 2 0 、 ブ ロ ッ ク 内 画素値設 定回 路 1 4 0 ,
1 5 0 、 動 き 補償予測 回 路 1 6 0 、 2 値画像符 号化 回 路 1 7 0 、 縮小 回 路 1 7 1 7 3 , 1 7 4 、 拡大 回 路 1 7 2 、 フ レ ー ム メ モ リ 1 3 0 、 転置 回 路 1 7 5 ,
1 7 6 、 ス キ ャ ン タ イ プ ( S Τ ) 判 定回 路 1 7 7 、 動 き べ ク ト ル検 出 回 路 ( Μ V Ε ) 1 7 8 、 Μ V 符 号化 回 路 1 7 9 、 V L C ( 可変長符 号化) · 多重化 回 路 1 8 0 力ゝ ら な る 。
こ れ ら の う ち 、 ブ ロ ッ ク 内 画素値設 定回 路 1 4 0 は マ ク ロ ブ 口 ッ ク 内 の 画素ィ直 を 全て透 明 と す る 画素 デー タ を発 生す る 回 路 で あ り 、 プ ロ ッ ク 内 画素値設 定回 路 1 5 0 はマ ク ロ ブ π ッ ク 内 の 画素値 を 全て 不透 明 と す る 画素デ— タ を発 生す る 回 路 で あ る 。
C R ( サイ ズ変換率) 判 定回 路 1 1 1 は 、 ア ル フ ァ マ ッ プ信 号入力線 2 を 介 し て 供給 さ れ る ア ル フ ァ マ ツ プ信号 を解析 し 、 1 フ レ ー ム 分 の ア ル フ ァ マ ッ プ画像 を ど の 程度 の サ イ ズ変換率 で処理すべ き か を 判 定す る と 共 に そ の判 定結果 を サ イ ズ変換率 ( convers ion ratio ) b 2 と し て 出 力 す る 。 ま た 、 縮小 回 路 1 7 1 は ア ル フ ァ マ ッ プ信号入力線 2 を 介 し て供給 さ れ る ァ ル フ ァ マ ッ プ信 号 を フ レ ー ム 全体 に つ レ、 て 、 ス キ ャ ン タ イ プ ( S T ) 判 定回 路 1 7 7 は 2 値画像符 号化 回 路 1 7 0 力ゝ ら の符 号化 出 力 を 基 に ス キ ャ ン方 向 を 判 定 し て ス キ ャ ン方 向 情報 ( scan type ) b 4 を 出 力 す る 。
ま た 、 転置 回 路 1 7 5 は こ の ス キ ャ ン タ イ プ ( S T ) 判 定回 路 1 7 7 の 出 力 す る ス キ ャ ン方 向 情報 b 4 を 基 に 、 縮小 回路 1 7 1 の縮小処理 し た 1 フ レ ー ム 分 の ァ ノレ フ ァ マ ッ プ信号 につ い て 各 マ ク ロ ブ 口 ッ ク の位 置 を 、 転置処理 し 、 転置回 路 1 7 6 は ス キ ャ ン タ イ プ 判 定回 路 1 7 7 の 出 力 す る ス キ ャ ン方 向 情報 b 4 を 基 に 、 縮小 回路 1 7 1 , 1 7 2 , 1 7 3 の 出 力 を 転置処 理 し て 出 力 す る 。 2 値画像符 号化 回 路 1 7 0 は こ れ ら の転置 回 路 1 7 5 、 1 7 6 を 介 し て 与 え ら れた縮小 ァ ノレ フ ァ マ ッ プ信号 を符 号化処理 し て 出 力 す る 。
ま た 、 拡大 回 路 1 7 2 は縮小 回 路 1 7 1 を 介 し て 与 え ら れ る ア ル フ ァ マ ッ プ信 号 を 前記 C R 判 定回 路 1 1 1 の 出 力 す る サ イ ズ変換率 で拡大処理 し 、 動 き 補償予 測 回路 1 6 0 は フ レ ー ム メ モ リ 1 3 0 に 蓄積 さ れて い る 参 照 フ レ ー ム の 再生画像 を 用 い て 動 き 補償予測信 号 を 生成 し 、 モ ー ド判 定回 路 1 1 0 お よ び縮小 回 路 1 7 4 に 出 力 す る 。 縮小 回 路 1 7 4 は動 き 補償予測信 号 を C R 判 定回 路 1 1 1 の 出 力 す る サ イ ズ変換率 で縮小処 理 し 、 縮小 回 路 1 7 3 は フ レ ー ム メ モ リ 1 3 0 に 蓄積 さ れて い る 参照 フ レ ー ム の 再生画像 を C R 判 定回 路 1 1 1 の 出 力す る サ イ ズ変換率で縮小処理す る 。
ま た 、 セ レ ク タ 1 2 0 はモ ー ド判 定回 路 1 1 0 の 出 力 す る 分類情報 に従 っ て プ ロ ッ ク 内 画素値設 定回 路 1 4 0 か ら の 再生信号 m 0 , プ ロ ッ ク 内 画素値設 定回路
1 5 0 力ゝ ら の 再生信号 m 1 , 動 き 補償予測 回 路 1 6 0 力 ら の 動 き 補償予測信 号 m 2 , 拡 大 回 路 1 7 2 力ゝ ら の 再生信 号 m 3 の う ち 、 所要 の 信 号 を 選択 し て 出 力 し 、 フ レ ー ム メ モ リ 1 3 0 は こ の セ レ ク タ 1 2 0 の 出 力 を フ レ ー ム 単位 で記憶す る 。
ま た 、 モ ー ド判 定回 路 1 1 0 は 、 動 き 補償予測 回路
1 6 0 か ら の 動 き 補償予測信 号 と 2 値画像符 号化 回 路 1 7 0 力、 ら の符 号ィ匕信 号 b 4 を 参 照 し て 、 ア ル フ ァ マ ッ プ信号入力線 2 を 介 し て 供給 さ れ る ア ル フ ァ マ ッ プ 信 号 を解析 し 、 マ ク ロ プ ロ ッ ク 毎 に
transparent 、 o aque 、 no update 、 " coded" の何れ に 分類 さ れ る カゝ を 判 定す る と 共 に判 定結果 を モ ー ド情報 b 0 と し て 出 力 す る 。
動 き ベ ク ト ル検 出 回 路 ( M V E ) 1 7 8 は 、 ア ル フ ァ マ ッ プ信 号入力線 2 を介 し て 供給 さ れ る ア ル フ ァ マ ッ プ信 号か ら 動 き べ ク ト ル を 検 出す る 。 M V 符 号化 回 路 1 7 9 は 、 動 き べ ク ト ル検 出 回 路 ( M V E ) 1 7 8 で検 出 さ れた動 き べ ク ト ル を 符 号化 し 、 そ の符 号化結 果 を 動 き べ ク ト ル情報 b 1 と し て 出 力 す る 。 例 え ば、 M V 符 号化 回路 1 7 9 に 予測符 号化 を 適用 し た 場合 に は そ の 予測誤差信 号が 動 き べ ク ト ル情報 b 1 と し て 出 力 さ れ る こ と に な る 。
V L C (可変長符 号化) · 多重化 回 路 1 8 0 は 、 モ 一 ド判 定回 路 1 1 0 力ゝ ら の モ ー ド情報 b 0 お よ び M V 符号化 回 路 1 7 9 力ゝ ら の 動 き べ ク ト ル情報 b 1 お よ び C R ( サ イ ズ変換率) 判 定回 路 1 1 1 か ら の サ イ ズ変 換率情報 b 2 お よ びス キ ャ ン タ イ プ ( S T ) 判 定回 路 1 7 7 力 ら の ス キ ャ ン方 向 情報 b 3 お よ び 2 値画像符 号化 回路 1 7 0 か ら の 2 値符 号化 情報 b 4 を 受 け て こ れ ら を 可変長符 号化 し 、 多重化 し て信号線 3 へ と 出 力 す る 。
こ の よ う な構成 に お い て 、 符 号化対象 で あ る ア ル フ ァ マ ッ プ信号は 、 ア ル フ ァ マ ッ プ信 号入力線 2 を 介 し て ア ル フ ァ マ ッ プ符 号化 回 路 2 0 に供給 さ れ る 。 こ れ を 受 け た ァ ル フ ァ マ ッ プ符 号化 回 路 2 0 は 、 そ の モ ー ド判 定回 路 1 1 0 に お レ、 て ア ル フ ァ マ ッ プ信 号 を解析 し 、 マ ク ロ フ ロ ッ ク '毋 に transparent 、
o p a σ u e 、 " no u date" 、 " coded" の 何れ に 分類 さ れ る か を 判 定す る 。 こ こ で 、 分類 の 評価基準 と し て は 、 例 え ば、 ミ ス マ ッ チ画素数 を 用 い る 。
具体的 に は次 の 通 り で あ る 。
モ ー ド判 定回 路 1 1 0 は 、 ま ず 、 入力 さ れた マ ク ロ ブ ロ ッ ク 内 の信 号 を 全て 透 明 に 置 き 換 え た場合 の ミ ス マ ッ チ画素数 を 計算 し 、 こ の 数が し き い値以 下 と な る マ ク ロ ブ ロ ッ ク を " transparent" に 分類す る 。 同 様 に 、 マ ク ロ ブ ロ ッ ク 内 を 全て 不透 明 に 置 き 換 え た場 合の ミ ス マ ッ チ画素数 が し き い値以 下 と な る マ ク ロ ブ ロ ッ ク o aque に 分 す 。
次 に 、 モ ー ド判 定回 路 1 1 0 は 、
transparent に も o aque (こ t> 分類 5 れな レヽ マ ク ロ ブ ロ ッ ク に 対 し て 、 信 号線 1 0 1 を 介 し て 供給 さ れ る 動 き 補償予測値 と の ミ ス マ ッ チ画素数 を 計算 し 、 こ の数力 S し き レヽ値以 下 と な る ブ ロ ッ ク を " no
update" に 分類す る 。
こ こ で 、 transparent o aque め る レヽ ま
" no update" に も 分類 さ れ な 力 つ た マ ク ロ ブ 口 ッ ク は 、 " coded" に 分類 さ れ る 。
モ ー ド判 定回 路 1 1 0 に よ る こ の 分類情報 b 0 は 、 信 号線 1 0 2 を 介 し て セ レ ク タ 1 2 0 に供給 さ れ、 セ レ ク タ 1 2 0 で は該 ブ ロ ッ ク の モ ー ド が
" transparent" の場合 は 、 ブ ロ ッ ク 内 画素ィ直設 定 回 路 1 4 0 に お い て プ ロ ッ ク 内 画素値が全て 透 明 に さ れた 再生信号 m O を 選択 し 、 信 号線 4 を 介 し て フ レ ー ム メ モ リ 1 3 0 に供給 し 、 該 フ レ ー ム の 記憶領域 に 蓄 え る と 共 に 、 ア ル フ ァ マ ッ プ符 号化 回 路 2 0 の 出 力 と し て 出 力 す る 。
同様 に 、 該マ ク ロ ブ ロ ッ ク の モ ー ド が " opaque" の 場合 は 、 セ レ ク タ 1 2 0 は マ ク ロ ブ ロ ッ ク 内 画素値 設 定回 路 1 5 0 に お い て マ ク ロ ブ 口 ッ ク 内 画素ィ直 が 全 て 不透 明 に さ れた 再生信 号 m 1 を 選択 し 、 ま た 、 該マ ク ロ ブ ロ ッ ク の モ ー ド が " no update" の 場合 は 、 信号線 1 0 1 を介 し て 供給 さ れ る 動 き 補償予測 回路 1 6 0 に よ り 生成 さ れた動 き 補償予測信 号 m 2 を 選択 し ま た 、 該マ ク p ブ ロ ッ ク の モ ー ド が " coded" の場合 に は 、 縮小 回 路 1 Ί 1 、 拡大 回 路 1 7 2 を経 て 供給 さ れ る 再生信 号 m 3 を 選択 し て 、 こ れ を 信 号線 4 を 介 し て フ レ ー ム メ モ リ 1 3 0 に供給 し 、 該 フ レ ー ム の 記憶 領域 に 蓄 え る と 共 に 、 ァ ル フ ァ マ ッ プ符 号化 回 路 2 0 の 出 力 信 号 と し て 出 力 す る 。
ま た 、 モ ー ド判 定回 路 1 1 0 に お レヽ て " coded" に 分類 さ れた マ ク ロ ブ ロ ッ ク の 画素値 は 、 縮小 回 路 1 7 1 に お い て 縮小 さ れた後 、 2 値画像符 号化 回 路 1 7 0 に お い て 符 号化 さ れ る 。 こ こ で 、 縮小 回 路 1 7 1 で用 い ら れ る サ イ ズ変換率 ( し ( Convers ion
Ratio ) ) の 設定情報 は 、 C R 判 定回路 1 1 1 に お い て 求 め ら れ る 。 例 え ば、 縮小率 を " 1 (縮小 し な い) ,, 、 " 1 / 2 (水 平 ' 垂直共 に 1 Z 2 ) " 、 " 1 / 4 (水 平 · 垂直共 に 1 / 4 ) " の 3 通 り に 定 め た場 合 、 C R 判 定回路 1 1 1 で は 、 次 の ス テ ッ プ で C R (サ イ ズ変換率) が 求 め ら れ る 。
( 1 ) 該 マ ク ロ ブ ロ ッ ク を " 1 4 ,, に縮小 し た 際 の 再生信 号 と 、 該マ ク ロ プ ロ ッ ク と の ミ ス マ ツ チ数 を 計算 し 、 こ の 数 が し き い値以 下 の 場合 に は 、 縮 小率 を " と す る 。
( 2 ) 前記 ( 1 ) で ミ ス マ ッ チ数が し き い値 よ り ち 大 き い場合 に は 、 該マ ク ロ ブ ロ ッ ク を " 1 Z 2 " に縮小 し た 際 の 再生信 号 と 、 該マ ク ロ ブ ロ ッ ク と の ミ ス マ ッ チ数 を 計算 し 、 こ の数 が し き い値以 下 の 楊 合 に は 、 縮小率 を " 1 / 2 " と す る 。
( 3 ) 前記 ( 2 ) で ミ ス マ ッ チ数が し き い値 ぶ り ち 大 き い場合 に は 、 該 マ ク ロ ブ ロ ッ ク の縮小率 を " 1 ,, と す る 。
こ の よ う に し て 求 め ら れた C R ( サ イ ズ変換率) の 値 は 、 信号線 1 0 3 を 介 し て 縮小 回 路 1 7 1 , 1 7 3 1 7 4 、 拡大 回 路 1 7 2 お よ び 2 値画像符 号化 回路 1 7 0 に供給 さ れ る と 共 に 、 V L C ( 可変長符号化) · 多重化 回 路 1 8 0 に供給 さ れ符 号化 さ れた後 、 他 の 符 号 と 多重化 さ れ る 。 ま た 、 転置 回 路 1 7 5 で は 、 縮小 回路 1 7 1 よ り 供給 さ れ る 、 縮小 さ れた該 ブ ロ ッ ク の 信 号の位置 が 転置 さ れ る (水 平方 向 の ア ド レ ス と 垂直 方 向 の ア ド レ ス が 入れ替 え ら れ る ) 。
こ の こ と に よ り 、 符 号化 の順序 を 水 平方 向 の ス キ ヤ ノ ( horizonta L s c a n )順 に行 う 力、 、 垂 直方 向 の ス キ ャ ン ( vertica s c a n )順 に行 う カゝ ; ^ 切 り 換 え ら れ る 。 ま た 、 転置 回 路 1 7 6 で は 、 縮小 回 路 1 7 3 に よ り 縮 小 さ れた該マ ク ロ ブ ロ ッ ク 近傍 の 再生画素値 と 、 縮小 回 路 1 7 4 に よ り 縮小 さ れた動 き 補償予測信 号 と で 、 そ れぞれの信 号の位置 が 転置 さ れ る 。 転置 回 路 1 7 5 , 1 7 6 に お レ、 て 、 転置処理す る か 否 かの 判 定は 、 例 え ば、 2 値画像符 号化 回 路 1 7 0 に お レヽ て 水 平ス キ ャ ン と 垂直 ス キ ャ ン と で そ れぞれ符 号 化 を行い 、 信号線 1 0 4 を 介 し て 出 力 さ れ る 符 号化情 報 b 4 を S T ( Scan Type) 判 定回 路 1 7 7 に供給 す る こ と に よ っ て 、 こ の S T 判 定回 路 1 7 7 力 S 、 符 号 量 の少 な く な る ス キ ャ ン方 向 を 選択す る こ と で決 ま る
2 値画像符号化 回 路 1 7 0 で は 、 転置 回 路 1 7 5 よ り 供給 さ れ る 該マ ク ロ ブ ロ ッ ク の 信 号 を 、 転置 回 路 1 7 6 よ り 供給 さ れ る 参 照信 号 を 用 い て 符 号化す る 。
な お 、 具体的 な 2 値画像符 号化法 の 一例 と し て 、 上 述 の 第 3 の 実施例 で用 い た 手法 が 挙 げ ら れ る が 、 こ れ に 限 ら ず 、 第 4 の 実施例 で の こ の 2 値画像符 号化 回 路 1 7 0 で は他の 2 値画像符 号化 を 適用 す る こ と が で き る 。
coded" に分類 さ れた ブ ロ ッ ク の 中 に は 、 フ レ ー ム 内 の 参 照画素 を 用 い る " intra" 符号化モ ー ド と 、 動 き 補償予測信号 を 参照す る " inter" 符 号化モ ー ド と が あ る 。 こ こ で 、 intra/ inter の 切 り 替 え は 、 例 え ば、 線 1 0 4 を 介 し て 供給 さ れ る 2 値画像符 号化 回 路 1 7 0 か ら の符 号化情報 を モ ー ド判 定回 路 1 1 0 に 供給 し 、 符号量 の少 な く な る モ ー ド を選択すれば よ い
、 _ ゝ で 、 選択 さ れた符 号ィ匕モ ー ド ( intra/ inter) の 情報 は 、 信号線 1 0 5 を 介 し て 2 値画像符 号化 回 路 1 7 0 に供給 さ れ る 。 以 上の 手段 に よ り 選択 さ れた最適 な モ ー ド に て 符 号 ィ匕 さ れた 、 2 値画像符 号化 回 路 1 7 0 の符 号化情報 は 、 信号線 1 0 4 を 介 し て V L C ' 多重化 回 路 1 8 0 に供 給 さ れ、 他の符号 と 共 に 多重化 さ れ る 。 ま た 、 モ ー ド 判 定回路 1 1 0 か ら は最適 な モ ー ド の 情報 が 信号線 1 0 6 を介 し て V L C · 多重化 回 路 1 8 0 に供給 さ れ、 符号化 さ れた 後 、 他 の符号 と 共 に 多重化 さ れ る 。 ま た 、 動 き ベ ク ト ル検 出 回 路 ( M V E ) 1 7 8 で は 、 最適 な 動 き べ ク ト ルが 検 出 さ れ る 。 こ こ で 、 動 き べ ク ト ルの 検 出 方法 につ い て は 、 種 々 存在 し て お り 、 ま た 、 動 き べ ク ト ルの検 出 法そ の も の は 、 本発 明 の 要部 で は な い た め 、 こ こ で は省略す る 。 検 出 さ れた動 き べ ク ト ル は 、 信 号線 1 0 7 を 介 し て 動 き 補償予測 回 路 1 6 0 に 供給 さ れ る と 共 に 、 M V 符 号化 回 路 1 7 9 に て 符 号化 さ れ た後 、 V L C · 多重化 回 路 1 8 0 に供給 さ れ、 符 号化 さ れた後 、 他 の 符号 と 共 に 多重化 さ れ る 。
多重化 さ れた符号 は 、 信 号線 3 を 介 し て 出 力 さ れ る 。 動 き 補償予測 回 路 1 6 0 で は 、 信号線 1 0 7 を 介 し て 供給 さ れ る 動 き べ ク ト ル信 号 に 基づい て フ レ ー ム メ モ リ 1 3 0 に 蓄積 さ れて い る 参 照 フ レ ー ム の 再生画像 を 用 い て 動 き 補償予測信号 を 生成 し 、 信 号線 1 0 1 を 介 し て モ ー ド判 定回 路 1 1 0 お よ び縮小 回 路 1 7 4 に 出 力 す る 。
以 上の 処理の 結果 、 画質 の 良 い 、 し か も 、 高圧縮率 で の 画像符 号化 を す る こ と が で き る 。 次 に復 号化 に っ レ、 て 説 明 す る 。
図 2 0 は 、 ア ル フ ァ マ ッ プ復 号化 回 路 4 0 の 詳細 な 構成 図 で あ る 。
図 に示す よ う に 、 ア ル フ ァ マ ッ プ復号化 回 路 4 0 は V L C (可変長符 号化) · 分離化 回 路 2 1 0 、 モ ー ド 再生回路 2 2 0 、 セ レ ク タ 2 3 0 、 マ ク ロ ブ ロ ッ ク 内 画素値設定回 路 2 4 0 , 2 5 0 、 動 き 補償予測 回 路 2 6 0 、 フ レ ー ム メ モ リ 2 7 0 、 2 値画像復 号化 回 路 2 8 0 、 拡大 回 路 2 8 1 、 転置 回 路 2 8 2 , 2 8 5 、 縮 小 回 路 2 8 3 , 2 8 4 、 2 値画像復 号化 回 路 2 8 0 、 動 き べ ク ト ル再生 回 路 2 9 0 よ り 構成 さ れ る 。
こ れ ら の う ち 、 V L C (可変長符 号化) · 分離化 回 路 2 1 0 は 、 多重化 さ れて 送 ら れて き た ア ル フ ァ マ ツ プ の符 号化 ビ ッ ト ス ト リ 一 ム を 復 号 し 、 モ ー ド情報 b 0 、 動 き べ ク ト ル情報 b 1 、 サ イ ズ変換率情報 b 2 、 ス キ ャ ン方 向 情報 b 3 、 2 値画像符 号化情報 b 4 に 分 離す る 回路 で あ り 、 モ ー ド再生回 路 2 2 0 は こ の 分離 さ れたモ ー ド情報 b 0 を 受 け て 、 " transparent" , o aque , no update , coded の 4 £ の モ ー ド の う ち の いずれの モ ー ド で あ る カゝ を 再生す る 回 路 で あ る 。
2 値画像復 号化 回 路 2 8 0 は前記 V L C ( 可変長符 号化) · 分離化 回 路 2 1 0 に よ り 分離 さ れた 2 値画像 符 号化情報 b 4 を 、 前記分離 さ れた サイ ズ変換率情報 b 2 お よ びモ ー ド再生 回 路 2 2 0 か ら の 再生 さ れた モ — ド情報お よ び転置 回 路 2 8 5 の 情報 と を 用 い て 2 値 画像 に復 号化 し て 出 力 す る 回 路 で あ り 、 拡大 回 路 2 8 1 は前記分離 さ れた サ イ ズ変換率情報 b 2 の 情報 を 用 い て 前記復 号 さ れた 2 値画像 を 拡大処理す る 回 路 で あ り 、 転置 回 路 2 8 2 は こ の 拡大処理 さ れた 画像 を 、 前 記分離 さ れた ス キ ャ ン方 向 情報 b 3 に従 っ て 転置処理 し 、 再生信 号 m 3 と し て 出 力 す る 回 路 で あ る 。
ま た 、 マ ク ロ プ ロ ッ ク 內 画素値設 定回 路 2 4 0 は , マ ク ロ ブ 口 ッ ク 内 画素値 が 全て 透 明 に さ れた 再生信 号 m O を発 生す る 回 路 で あ り 、 マ ク ロ ブ ロ ッ ク 内 画素値 設 定回路 2 5 0 は, マ ク ロ ブ ロ ッ ク 内 画素値 が全て 不 透 明 に さ れた再生信 号 m 1 を発 生す る 回 路 で あ る 。
動 き ベ ク ト ル再生 回 路 2 9 0 は 、 前記 V L C ( 可変 長符 号化) · 分離化 回 路 2 1 0 に よ り 分離 さ れた動 き ベ タ ト ル情報 b 1 を 用 レ、 て マ ク ロ プ ロ ッ ク の 動 き べ ク ト ル を 再生す る 回 路 で あ り 、 動 き 補償予測 回 路 2 6 0 は こ の 再生 さ れた動 き べ ク ト ル を 用 い て フ レ ー ム メ モ リ 2 7 0 の 蓄積参照 フ レ ー ム を 再生画像 か ら 動 き 補償 予測値 m 2 を 生成す る 回 路 で あ り 、 縮小 回 路 2 8 4 は 前記分離 さ れた サ イ ズ変換率情報 b 2 の 情報 を 用 い て 前記動 き 補償予測値 m 2 を縮小処理す る も の で あ り 、 縮小 回 路 2 8 3 は前記分離 さ れた サ イ ズ変換率情報 b 2 の 情報 を 用 い て 前記 フ レ ー ム メ モ リ 2 7 0 の 蓄積参 照 フ レ ー ム の 画像 を縮小処理す る も の で あ り 、 転置 回 路 2 8 5 は前記 V L C ( 可変長符 号化) ' 分離化 回 路 2 1 0 に よ り 分離 さ れた ス キ ャ ン方 向 情報 b 3 を 基 に 縮小 回 路 2 8 3 , 2 8 4 の 出 力 す る 縮小画像 を 転置処 理 し て 2 値画像復 号化 回 路 2 8 0 に 出 力 す る 回 路 で あ る 。
ま た 、 セ レ ク タ 2 3 0 は 、 モ ー ド再生 回 路 2 2 0 の 出 力 す る モ ー ド情報 に応 じ て 、 マ ク 口 プ ロ ッ ク 内 画素 値設 定回 路 2 4 0 , 2 5 0 の 再生信 号 m O , m l 、 動 き 補償予測 回 路 2 6 0 の動 き 補償予測値 m 2 、 転置 回 路 2 8 2 の 再生信 号 m 3 の う ち 、 いずれか を 選択 し て 出 力 す る も の で あ り 、 フ レ ー ム メ モ リ 2 7 0 は こ の セ レ ク タ 2 3 0 の 出 力 し た 画像信 号 を フ レ ー ム 単位 で保 持す る メ モ リ で あ る 。
こ の よ う な構成 の ア ル フ ァ マ ッ プ複 号化 回 路 4 0 は 信 号線 8 を 介 し て ア ル フ ァ マ ッ プ の符 号化 ビ ッ ト ス ト リ ー ム が供給 さ れ る 。 こ の ビ ッ ト ス ト リ ー ム は V L D ( 可変長復 号化) · 分離化回 路 2 1 0 に供給 さ れ る 。 す る と 、 V L D (可変長複 号化) · 分離化 回 路 2 1 0 で は 、 こ の ビ ッ ト ス ト リ ー ム を 復 号 し 、 力 つ 、 モ ー ド情報 b 0 、 動 き べ ク ト ル情報 b 1 、 サ イ ズ変換率情 報 b 2 、 ス キ ャ ン方 向 情報 b 3 、 2 値画像符 号化情報 b 4 に 分離す る 。 こ れ ら 分離 さ れた 情報 は マ ク ロ プ ロ ッ ク 単位 で管理 さ れ る 。
こ の 分離 さ れた 情報 の う ち 、 モ ー ド情報 b 0 は 、 モ ー ド再生回 跨 2 2 0 に供給 さ れ、 該マ ク ロ プ ロ ッ ク が 以 下 の モ一 ド の 何れ に 属す る か を 分類す る 。 ( 1 ) transparent"
( 2 ) opaque"
( 3 ) " no u date"
( 4 ) " coded"
こ こ で 、 " coded" に は 、 図 1 9 の符 号化 回 路 の 実 施例 で述べた と お り " i n t r a "の モ ー ド と " inter" の モ ー ド と が あ る 。 す な わ ち 、 フ レ ー ム 内 の 参 照 画素 を用 レ、 る の が " i n t r a "符 号ィ匕モ ー ド で あ り 、 動 き 補 償予測信 号 を 参 照す る の が " inter" 符 号化モ ー ド で あ る 。
セ レ ク タ 2 3 0 で は 、 信 号線 2 0 1 を 介 し て 供給 さ れ る 前記の符号化モ ー ド に応 じ て 、 該マ ク ロ ブ 口 ッ ク の モ ー ド が " transparent" の 場合 は 、 ブ ロ ッ ク 内 画素値設 定回路 2 4 0 に お い て マ ク 口 プ ロ ッ ク 内 画素 値が全て 透明 に さ れた 再生信 号 m 0 が 選択 さ れ、 信 号 線 9 を 介 し て フ レ ー ム メ モ リ 2 7 0 に供給 さ れ 、 該 フ レ ー ム メ モ リ 2 7 0 の 当 該マ ク ロ プ ロ ッ ク カ S 属 す る 画 像 フ レ ー ム の 記憶領域 に 蓄 え ら れ る と 共 に 、 ア ル フ ァ マ ッ プ複号化 回 路 4 0 か ら 復 号化 さ れた ア ル フ ァ マ ツ プの 画像 と し て 出 力 さ れ る 。
同様 に 、 該マ ク ロ ブ ロ ッ ク の モ ー ド が " opaque" の 場合 は 、 マ ク ロ ブ ロ ッ ク 内 画素値設 定回 路 2 5 0 に お い て 、 プ ロ ッ ク 内 画素値が 全て 不透明 に さ れた 再生 信号 m 1 が選択 さ れ、 ま た 、 該 ブ ロ ッ ク の モ ー ド が " no update" の 場合 は 、 信 号線 2 0 2 を 介 し て 供 給 さ れ る 動 き 補償予測 回 路 2 6 0 に お い て 生成 さ れ た 動 き 補償予測信 号 m 2 が 選択 さ れ、 ま た 、 該 ブ ロ ッ ク の モ ー ド が 、、 " coded" の 場合 に は 、 拡大 回 路 2 8 1 、 転置 回路 2 8 2 を 経 て 供給 さ れ る 再生信 号 m 3 が 選択 さ れ、 信 号線 9 を介 し て フ レ ー ム メ モ リ 2 7 0 に供給 さ れ、 該 フ レ ー ム メ モ リ 2 7 0 の 当 該マ ク ロ ブ ロ ッ ク が 属す る 画像 フ レ ー ム の 記憶領域 に 蓄 え ら れ る と 共 に 、 ア ル フ ァ マ ッ プ復 号化 回 路 4 0 か ら 復 号化 さ れた ア ル フ ァ マ ッ プ の 画像 と し て 出 力 さ れ る 。
一方 、 分離 さ れた動 き べ ク ト ル情報 b 1 は 、 動 き べ ク ト ル再生 回 路 2 9 0 に供給 さ れ、 該マ ク ロ ブ ロ ッ ク の 動 き べ ク ト ルが 再生 さ れ る 。 こ こ で 、 再生 さ れた 動 き べ ク ト ルは 、 信 号線 2 0 3 を 介 し て 動 き 補償予測 回 路 2 6 0 に供給 さ れ、 動 き 補償予測 回 路 2 6 0 で は こ の 動 き ベ ク ト ル に 基 づ レ、 て 、 フ レ ー ム メ モ リ 2 7 0 に 蓄積 さ れて い る 参 照 フ レ ー ム の 再生 面像 か ら 動 き 補償 予測値 を 生成す る 。 信 号線 2 0 2 を介 し て縮小 回 路 2 8 4 と セ レ ク タ 2 3 0 と に 出 力す る 。
ま た 、 分離 さ れた サ イ ズ変換率情報 b 2 は 、 縮小 回 路 2 8 3 お よ び拡大 回 路 2 8 1 に供給 さ れ る と 共 に 、 2 値画像復 号化 回 路 2 8 0 に供給 さ れ、 ス キ ャ ン方 向 情報 b 3 は 、 転置 回 路 2 8 2 , 2 8 5 に供給 さ れ る 。
ス キ ャ ン方 向 情報 b 3 を 受 け た転置 回 路 2 8 2 で は 、 拡大 回 路 2 8 1 よ り 供給 さ れ る 、 元 の サ イ ズ に 戻 さ れ た マ ク ロ ブ ロ ッ ク の 再生信 号の位置 が転置 さ れ る (水 平方 向 の ァ ド レ ス と 垂直方 向 の ァ ド レ ス が 入れ替 え ら れ る ) 。 ま た 、 ス キ ャ ン方 向 情報 b 3 を 受 け た転置 回 路 2 8 5 で は 、 縮小 回路 2 8 3 に よ り 縮小 さ れた該ブ ロ ッ ク 近傍の 再生画素値 と 、 縮小 回 路 2 8 4 に よ り 縮 小 さ れた動 き 補償予測信 号 と で 、 そ れぞれの 信 号の位 置 が転置 さ れ る 。
サ イ ズ変換率情報 b 2 を 受 け た 2 値画像複 号化 回 路 2 8 0 で は 、 分離化 回 路 2 1 0 に よ り 分離 さ れた該マ ク ロ プ ロ ッ ク の 2 値画像符 号化情報 b 4 を 、 転置 回 路 2 8 5 よ り 供給 さ れ る 参照信 号 を 用 い て 復 号化す る 。
と こ ろ で 、 第 4 の 実施例 に記載 し た よ う に 、 拡大 回 路 1 7 2 , 2 8 1 の 出 力 に は斜 め方 向 の 不連続性 に起 因す る 画質劣化 が発 生す る 場合 が あ る が 、 こ の 問題 を 解決す る た め に拡大 回 路 1 7 2 , 2 8 1 に は 、 斜 め方 向 の 不連続性 を抑 え る フ ィ ル タ を備 え る よ う に し て も 良 い。
な お 、 2 値画像符 号化 回 路 1 7 0 と 同様 に 、 用 レ、 る 2 値画像複号化 回 路 2 8 0 の構成 は 上述 の 第 3 の 実施 例 に示 し た構成例 に 限 ら な い。
前記 の 実施例 の 中 で 、 2 値画像符 号化 回 路 1 7 0 お よ び 2 値画像復 号化 回 路 2 8 0 の 実施例 は 、 上述 の 第 3 の 実施例 に 限 ら な レ、 と 記述 し た。 そ こ で 、 こ れ以外 の例 を 示 し て お く 。
別 の 2 値画像符 号化法 と し て 、 例 え ば、 マ ル コ フ モ デル符 号化法 が あ る (参照 : テ レ ビ ジ ョ ン学会編 、 "画像情報圧縮 " 、 pp · 171 〜 176 ) 。 図 2 1 が マ ル コ フ モ デル符 号化法 の 一例 を説 明 す る 図 で あ る 。 図 中 の 画素 X が符 号化 の 対象 と な る 着 目 画素 で あ り 、 画 素 a ~ f が 画素 X を符 号化す る 際 に 参照 さ れ る 参 照 画 素 で あ る 。
こ こ で 、 画素 a 〜 f は画素 X を 符 号化す る 際 に既 に 符 号化 が 終了 し て い る 画素 と す る 。 着 目 画素 X は 、 参 照 画素 a 〜 f の 状態 に よ っ て 、 V L C ( 可変長符 号 化) を使用 す る 場合 に は 可変長符 号テ 一 ブルが 、 算 術 符 号 ( arithmetic ) を使用 す る 場合 に は確率テ ー プ ノレ ( probability table )力 S 、 適応 的 に 切 り 換 え ら れ て 符号化 さ れ る 。
こ の 実施例 を本発 明 に適用 す る 場合 、 " intra" 符 号化 は 、 現在処理 し よ う と し て い る マ ク ロ プ ロ ッ ク 内 で 、 着 目 画素 よ り 先 に符 号化 さ れた 画素 を 参 照 画素 と し 、 " inter" 符 号化 は現在処理 し よ う と し て い る マ ク ロ ブ 口 ッ ク 内 で 、 着 目 画素 よ り 先 に符 号化 さ れた 画 素 だ け で な く 、 動 き 補償予測誤差信 号内 の 画素 も 参照 画素 と し て 用 いれ ば よ い。
本実施例 で は 、 例 え ば 、 [方式 1 ] 第 3 の 実施例 の 方式 ( M M R に基づ い た符 号化方式) と 、 [方式 2 ] マ ル コ フ モ デル符 号化 を 適応的 に切 り 換 え て 符 号化す る 方式の 両方式 を備 え 、 こ れ ら を 適応的 に選択 し て使 用 す る 形態 を採用 す る こ と を 特徴 と し て レ、 る 。
図 2 2 A は 、 本実施例 で用 い ら れ る 、 2 値画像符 号 化 回 路 1 7 0 の ブ ロ ッ ク 図 で あ る 。 図 に示す よ う に 、 2 値画像符 号化 回 路 1 7 0 は 、 一対 の セ レ ク タ 4 2 1 , 4 2 4 と 、 一対の 2 値画像符 号化 回 路 か ら 構成 さ れて い る 。 こ れ ら の う ち 、 セ レ ク タ 4 2 1 は入力 側 の 選択 回 路 で あ り 、 セ レ ク タ 4 2 2 は 出 力側 の 選択 回 路 で あ る 。 2 値画像符号化 回 路 4 2 2 は前記 [方式 1 ] で符 号化す る 第 1 の 2 値画像符 号化 回 路 で あ り 、 2 値画像 符 号化 回 路 4 2 3 は前記 [方式 2 ] で符 号化す る 第 2 の 2 値画像符 号化 回 路 で あ る 。
2 値画像符 号化 回 路 1 7 0 で は 、 信 号線 4 1 1 を 介 し て 転置 回 路 1 7 5 か ら 供給 さ れ る 処理対象 マ ク ロ ブ ロ ッ ク の 信 号 を 、 信 号線 4 1 3 を 介 し て 供給 さ れ る 切 り 換 え 信 号 に応 じ て 選択切換 え す る セ レ ク タ 4 2 1 , 4 2 4 に よ り 、 第 1 の 2 値画像符 号化 回 路 4 2 2 と 第 2 の 2 値画像符号化 回 路 4 2 3 に適応 的 に振 り 分 け 、 こ の 振 り 分 け ら れた マ ク ロ ブ 口 ッ ク の信 号 を 符 号化す る 。
符 号化 さ れた情報 は 、 信 号線 4 1 2 を 介 し て 出 力 さ れ る 。 な お 、 図 2 2 A で は 2 値画像符 号化 回 路 1 7 0 に供給 さ れ る 信号の信 号線 1 0 3 , 1 0 5 、 転置 回 路 1 7 6 力ゝ ら の信 号線は 、 省略 さ れて い る 。
こ こ で 、 信 号線 4 1 3 を 介 し て 供給 さ れ る 切 り 換 え 情報 は 、 予 め設定 さ れた デ フ ォ ル ト 値で も 良 い し 、 画 像 内 容 に 基づい て 適切 な 切 り 換 え 情報 を 求 め て 、 サ イ ド情報 と し て 別途符 号化 し 、 復 号側 に伝送す る よ う に し て も よ レヽ 。
こ れ に よ り 、 画像 内 容 に 基づ い た最適処理 が 可能 に な り 、 ま た 、 ア プ リ ケ ー シ ョ ン に応 じ て複数の符 号化 方式の 中 か ら 適切 な も の を 選択 し て使用 す る こ と も 可 能 に な る 。
同様 に 、 図 2 2 A は 、 本実施例 で用 い ら れ る 2 値画 像複 号化 回路 2 8 0 の プ ロ ッ ク 図 で あ る 。 図 に示す よ う に 、 2 値画像複 号化 回 路 2 8 0 は一 対 の セ レ ク タ 4 4 1 , 4 4 4 、 一対 の 2 値画像複 号化 回 路 4 4 2 , 4 4 3 と 力 ら 構成 さ れて レ、 る 。 こ れ ら の う ち 、 セ レ ク タ
4 4 1 は入力側 の 選択 回 路 で あ り 、 セ レ ク タ 4 4 4 は 出 力側 の選択回 路 で あ る 。 ま た 、 2 値画像複 号化 回 路 4 4 2 は前記 [方式 1 ] で符 号化 し た 画像 を 復 号化す る 第 1 の 2 値画像複 号化 回 路 で あ り 、 2 値画像複 号化 回路 4 4 3 は前記 [方式 2 ] 符 号化 し た 画像 を 復 号す る 第 2 の 2 値画像複 号化 回 路 で あ る 。
こ の よ う な構成 の 2 値画像複 号化 回 路 2 8 0 で は 、 信 号線 4 3 1 を 介 し て 供絵 さ れ る 処理対象 マ ク ロ ブ ロ ッ ク の 2 値画像符 号化情報 b 4 を 、 信 号線 4 3 3 を 介 し て供給 さ れ る 切 り 換 え 信 号 に応 じ て セ レ ク タ 4 4 1 , 4 4 4 に よ り 、 第 1 の 2 値画像復 号化 回 路 4 4 2 と 第 2 の 2 値画像復 号化 回 路 4 4 3 に適応 的 に振 り 分 け て 複号化す る 。
復号化 さ れた信 号は 、 信 号線 4 3 2 を 介 し て 出 力 さ れ る 。 な お 、 図 2 2 B で は 2 値画像複 号化 回 路 2 8 0 に対す る 分離化 回 路 2 1 0 か ら の信 号供給用 の信 号線 モ ー ド再生 回 路 2 2 0 か ら の信 号線, 転置 回 路 2 8 5 力、 ら の 信 号線は省略 さ れて い る 。
こ こ で 、 信 号線 4 3 3 を 介 し て 供給 さ れ る 切 り 換 え 情報 は 、 デ フ ォ ル ト 値 で も 良 い し 、 符 号化 さ れて 送 ら れて き た ビ ッ ト ス ト リ 一 ム かち 当 該切換情報 を 得 る ェ ン コ ー ダカゝ ら 送 ら れて き た 情報 で も 良 い。
次 に 、 サイ ズ変換 (拡大 , 縮小) を 行 う た め の 回 路 例 を 具体的 に示す。
本発 明 は 、 符号ィ匕領域 (video ob ect plane
( V 0 P ) )単位 あ る レ、 は プ ロ ッ ク 単位 ( マ ク ロ ブ 口 ッ ク ( macro block )単位) でサ イ ズ変換処理 を 施す こ と で レー ト 制御 ( rate control )を行 っ て レヽ る 。 こ の サ ィ ズ変換処理 に利用 す る 技術の例 と し て 、 本願発 明 者 等 が 特願平 8 — 2 3 7 0 5 3 号 に 開 示 し た "双 一次 内 挿 " が あ る 。 こ の 技術 に つ い て 触れて お く 。 当 該双一 次 内挿 は 、 参考文献 "尾 上編 : 画像処理ハ ン ド プ ッ ク , p . 6 3 0 , 昭晃 堂 " を 図 2 3 A お よ び 2 3 B を 参 照 し て 説明 す る 。
図 2 3 A に お い て 、 P e X は変換後 の 画素位置 で あ り 、 こ の P ex は 図 2 3 A の よ う に 実数画素位置 を 指 し示す。
そ こ で 、 入力 信 号の整数画素位置 A , B , C , D と の 距離関係 力ゝ ら 、 8 つ の領域 に 分 け て 図 2 3 B に示す 論理式 に よ り 、 A 〜 D の 画素値 I a 〜 I d 力 ら P ex の 画素ィ直 I p を 求 め る 。
こ の よ う な 処理 が "双 一 次 内挿 " と い う 処理 で あ り A 〜 D の 画素 直 I a 〜 I d 力、 ら P e x の 画素 直 I ρ を 簡易 に 求 め る こ と が で き る 。
と こ ろ で 、 こ の双一次 内揷処理 は周 囲 の 4 画素 の み を 用 い て 処理 を 行 っ て い る た め 、 広 レヽ範囲 で の 変化 が 反 映 さ れず 、 特 に斜 め 方 向 の 不連続性 が発 生 し易 く な つ て い る 。 こ の 問題 を解決す る た め の 一例 と し て 、 拡 大処理 を 行 っ た 後 に ス ム ー シ ン グ フ ィ ル タ 処理 ( 平 滑 化処理) を施す例 を 、 特願平 8 — 2 3 7 0 5 3 号で提 示 し て レヽ る 。
ス ム ー シ ン グ処理 (平滑化処理) を 図 2 4 A 、 図 2 4 B を 参 照 し て 具体的 に 説 明 す る 。 図 2 4 A は元 の サ ィ ズの 2 値画像 、 図 2 4 B は こ れ を 縮小 し て 得 た 2 値 画像 で あ る 。 同 図 に お い て は 、 オブ ジ ェ ク ト 領域 は黒 丸 印 で 、 ま た 、 ノ ッ ク グ ラ ウ ン ド (背 景) 領域 は 白 丸 印 で示 し て あ る 。
こ の例 で は 、 サ ン プ リ ン グ変換 (拡大 · 縮小変換) が な さ れ る こ と に よ り 発 生す る 斜 め 方 向 の 不連続性 を 滑 ら か にす る た め に 、 ノ ッ ク グ ラ ウ ン ド領域 の 画素
( 白 丸) 一 つ 一 つ に つ い て 、 そ れ を 中 心 に し て 、 そ の 上下左右 の 画素 、 つ ま り 、 隣接画素 を 調べ 、 そ の う ち 2 画素以 上 が オブ ジ ェ ク ト 領域の 画素 (黒丸) で あ つ た と き は 、 そ のバ ッ ク グ ラ ウ ン ド領域 の 画素 を 、 ォブ ジ ェ ク ト 領域 に含 め る 処理 を行 う 。 す な わ ち 、 今 、 ノ ッ ク グ ラ ウ ン ド領域 に あ る 一つ の 画素 で あ る 検査対象画素 が ず 2 4 B に お け る 二重丸 印 で示す位置 の画素 で あ る 場合 の よ う に 、 そ の 隣接画素 に 、 2 画素以上 、 オブ ジ ェ ク ト 領域 の 画素 (黒丸) が あ つ た と き は 、 そ の 二重丸 印 で示す位置 の 画素 (つ ま り 、 検査対象画素 ) を 黒丸印 の 画素 に し て オブ ジ ェ ク ト 領域 の 画素 にす る 。 黒丸 印 の 画素 が例 え ば 、 " 1 " . 白 丸印 が " 0 " で あ る と す る と 、 二重丸 印 で示す位置 の 画素 ( 画素値 " 0 " ) を 、 画素値 " 1 " に 置 き 換 え る 処理 を す る 。 こ の よ う な 処理 に よ り 、 斜 め 方 向 の 不 連続性 を 解消 で き る よ う に な る 。
図 2 5 は 、 ス ム ー シ ン グ フ ィ ル タ (平滑化処理 フ ィ ル タ ) の 別 の例 で あ る 。 図 2 5 に示 さ れ る 、 3 X 3 画 素 の マ ス ク の 中 心 の 画素 を C と し て 、 C に 対 し て 左 上 の 画素 を T L 、 右 上 の 画素 を T R 、 左 下 の 画素 を B L , 右 下 の 画素 を B R と す る と 、 下記 の 式 に よ り 、 画素 C の フ ィ ル タ リ ン グ さ れた値が 求 め ら れ る 。 こ こ で 、 ォ ブ ジ ェ ク ト の値 を " 1 " 、 背景の値 を " 0 " で表 わ し て レ、 る 。
i f ( C = = 0 ) i f ( ( T L + T R + B L + B R ) > 2 )
C = 1 :
e l s e
C = 0 : }
e l s e
{
i f ( ( T L + T R + B L + B R ) < 2 ) C = 1 ··
e l s e C
= 0 す な わ ち 、 こ の フ ィ ル タ 演算 処理 は 、 も し 画素 C が " 0 " で あれば、 画素 T L, T R, B L , B R をカロ 算 し た値が " 2 " よ り 大 き レ、 か否カゝ を 調べ 、 大 き い場合 は画素 C の値 を " 1 " と し 、 大 き く な い場合 は 画素 C を " 0 " と し 、 ま た 、 も し 画素 C 力 S " 0 " で な け れば 画素 T L, T R, B L, B R を力 B算 し た値が " 2 " よ り 小 さ い か否カゝ を 調 べ 、 小 さ い場合 は画素 C の値 を
" 1 " と し 、 大 き く な い場合 は画素 C を " 0 " と す る 処理で あ る 。
こ の フ ィ ル タ に よ れば、 対象 と な る 画素 C に 対 し て 斜 め方 向 に位置す る 画素値の 変化 を 考慮 し て 画素 cの 値 を修正す る た め 、 斜 め 方 向 の 不連続性 が解 消 さ れ る , な お 、 ス ム ー シ ン グ フ ィ ル タ の構成 は前記例 に 限 つ た も の で は な く 、 メ デ ィ ア ン フ ィ ル タ (medium
f ilter )等 の 非線开 フ ィ ノレ タ ( nonlinear f ilter ) を 用 い て も 良 レヽ 。
図 2 6 A は 、 双一 次 内 挿 に よ り 水 平 ' 垂 直共 に 2 倍 に拡大す る 処理を表す図 で あ る 。 図 2 6 B におい て 、 縮小プ ロ ッ ク の画素 ( decoded pixel )は 白 丸印すな わ ち " o " 印 で表 さ れ、 内挿 さ れ る 画素
( interpo l al ed p i X e 1 )はノ ッ 印すな わ ち 、 " X ,, 印 で表 さ れて お り 、 それぞれの画素値は " 0 " あ る い は " 1 " の何れかで あ る 。
こ の場合、 図 2 3 B の論理式で内挿 さ れる 画素 ( interpo l al ed pixel )の値を求 め る と 、
I p 1 = I a I ρ 2 = I b , I D 3 = I c , I p 4 = I d
と な る 。
従 っ て 、 画素 A の周 り の 4 つの 内挿画素の画素 ^ Sは 全て I a と な る た め に 、 拡大画像は 2 X 2 画素が 同 じ 値 と な り 、 滑 ら か さ が失われる 。 そ こ で、 双一次 内挿 の重み付 け を以下の様 に変更す る こ と で、 前記の 問題 は解決 さ れる
Ip l: i f ( 2 *: : a + I b + I c + I d > 2 ) then 1 " e l s e " 0 "
p 2 : i f ( I a + 2 * I b + I c + I d > 2 ) t he n e l s e 0 "
I p 3 : i f ( I a + I b + 2 * I c + I d > 2 ) t hen " " e l s e " 0 "
I p 4 : i f ( I a + I b + I c + 2 * I d > 2 ) t hen " " el s e " 0 "
こ こ で 、 P i ( i = l , 2 , 3 , 4 ) は、 図 2 6 A に示 さ れ る " X " 位置 に 対応 し た 内 挿 さ れ る 画素 で あ り 、 I p i ( i = 1 , 2 , 3 , 4 ) は画素 P i ( i = 1 , 2 , 3 , 4 ) の 画素値 ( " 1 " ま た は " 0 " ) で あ る 。 ま た 、 A, B , C, D は縮小 ブ ロ ッ ク の 画素 、 I a , I b , I c , I d は こ れ ら 画素 A , B , C,
D の 画素値で あ る 。
『 Ipl s if ( 2 * I a + I b + I c + I d > 2 ) t he n
" l " else " 0 "』 の 意味 は 、 画素 ρ 1 の 画素値 I pi は 、 も し 、 l a の 2 倍値 と 、 l b と 、 I c と 、 I d と の 和 力 S 2 よ り 大 き け れば " 1 " と し 、 そ う で な け れ ば " 0 " と す る と い う こ と を 示 し て お り 、
『 Ip2 : if ( Ia + 2 * I b + I c + I d > 2 ) t hen
" 1 " e 1 s e " 0 "』 の 意味 は 、
画素 p 2 の 画素値 I p2 は 、 も し 、 l b の 2 倍値 と 、 l a と 、 I c と 、 I d と の 和 力 2 よ り 大 き け れ ば、 " 1 " と し 、 そ う で な け れば " 0 " と す る と 云 う こ と を 示 し て お り 、
『 Ip3 = if ( Ia + I b +2 * Ic+Id > 2 ) then " l " else " 0 "』 の 意味 は 、
画素 ρ 3 の 画素ィ直 I p3 は 、 も し 、 I c の 2 倍ィ直 と 、 l a と 、 l b と 、 I d と の 和 力 S 2 よ り 大 き け れ ば、 " 1 " と し 、 そ う で な ければ " 0 " と す る と 云 う こ と を 示 し て お り 、
『 Ip4 s if ( Ia + I b + I c + 2 * I d > 2 ) t hen
" 1 " else " 0 "』 の意味 は 、 画素 p 4 の 画素値 I p 4 は 、 も し 、 I d の 2 倍値 と 、 I a と 、 I b と 、 I c と の 和 が 2 以 上 な ら ば、 " 1 " と し 、 そ う で な け れば " 0 " と す る と 云 う こ と を 示 し て レヽ る 。
な お 、 水 平 垂直共 に 4 倍 に 拡大す る 場合 は 、 前記 の 処理 を 2 回繰 り 返せ ば よ い。
以 上 は 、 演算 に よ り 拡大処理す る 例 で あ る が 、 演算 に よ ら ず と も 拡大処理す る こ と は で き る 。 そ の 例 を 次 に説 明 す る 。
こ こ で は 、 テ 一 ブル を 用 意 し て こ れ を メ モ リ に保持 さ せ 、 こ の テ 一 ブル に従 っ て 一義的 に 画素 を 置 き 換 え る よ う にす る も の で あ る 。
具体的 に説明 す る 。 ま ず 、 例 え ば メ モ リ の ア ド レ ス を 4 ビ ッ ト と し 、 I a , I b , I c , I d を 並べて 得 ら れ る ア ド レ ス に 、 そ の 、 I a , I b , I c , I d の ノヽ。 タ ー ン に よ つ て 得 ら れ る I pi, I p2 , I p3, I p4 を 下表 の様 に: 予 め 記録 し て お く 。
I a l b I c I d I pi I p 2 I p 3 I
P
0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0
0 0 1 0 0 0 0 0
0 0 1 1 0 0 1 1
0 1 0 0 0 0 0 0
0 1 0 1 0 1 0 1 0 1 1 0 0 1 1 0
0 1 1 1 1 1 1 1
1 0 0 0 0 0 0 0
1 0 0 1 1 0 0 1
1 0 1 0 1 0 1 0
1 0 1 1 1 1 1 1
1 1 0 0 1 1 0 0
1 1 0 1 1 1 1 1
1 1 1 0 1 1 1 1
1 1 1 1 1 1 1 1 こ の例 は 、 " I a , I b , I c , I d ,, 力 S " 0 , 0
0 , 0 " の 時 は 、 " I p 1, I p 2, I p 3, I p 4,, カ " 0 , 0 , 0 , 0 " で あ る こ と を 、 " I a , I b , I c , I d " が 0 , 0 , 0 , 1 ,, の 時 は 、 " I pi,
1 p 2 , I p 3, I p 4,, カ " 0 , 0 , 0 , 0 " で あ る こ と を 、 " I a 1 b , I c , I d ,, 力" 0 , 0 , 1 ,
1 ,, の 時 は 、 I pi, I p2, I p 3 , Ι ρ4Ί^ 0 ,
0 , 1 , 1 ,, で あ る こ と を 、 " I a , I b , I c , I d ,, が " 0 , 1 , 0 , 1 ,, の 時 は 、 " I p 1 , I p2 ,
1 p 3 , I p4" 力 S " 0 , 1 , 0 , 1 " で あ る こ と を 、 そ し て 、 " l a , I b , I c , I d ,, 力 " 0 , 1 , 1 0 ,, の 時 は 、 " I p 1 , I p 2 , I p 3, I p 4,, 力 S " 0 ,
, 0 " で あ る こ と を 、 … と 云 う 具合 に I a , I b , I c , I d の 内 容 の組み合わせ が 決 ま れば、 I p 1 , I p 2, I p 3 , I p 4 が 一義的 に 定 ま る テ ー ブル を 表 し て レヽ る 。
こ の よ う な テ 一 ブノレ を l a , I b , I c , I d の 內 容 が ア ド レ ス 、 そ の ア ド レ ス の格納デー タ が I pl, I p2, I p3, I p4 と な る よ う に メ モ リ に 設 定保持 さ せ手お く と 、 内 揷処理 を 行 う 時 に は 、 メ モ リ に l a I b , I c , I d の ア ド レ ス を 入力 し て 、 そ れ に 対応 す る I p 1, I p 2 , I p 3, I p 4 を 読み 出 す だ け で 、 内 揷値 を 求 め る こ と 力 S で き る 。 こ こ で 、 I a, I b, I c , I d の 並び は 2 進数で あ る が 、 こ れ を 1 0 進数 に 変換 し た数値 を コ ン テ キ ス ト と 呼ぶ こ と と す る と 、 こ の 手法 は 、 I a, I b, I c, I d に よ っ て 求 ま る コ ンテ キ ス ト に よ っ て 、 内 揷値 I pi, I p2 , I p3,
1 p4 を 決 め る 実施例 と 言 え る 。
な お 、 コ ン テ キ ス ト は
コ ン テ キ ス ト = 2 * 2 * 2 * 1 a + 2 * 2 * I b +
2 * I c + I d
で求 ま る 。
コ ン テ キ ス ト に よ っ て 内 揷値 を 求 め る 手法 は 、 上述 の例 の 場合 、 参照す る 画素数 を 4 つ と し た が 、 こ れ に 限 ら ず 、 以 下 で説 明 す る 1 2 個等 、 い く つ の 場合 で も 実現で き る 。
ま た 、 参 照 画素 ( decoded pixel ) と 内 ί荦画素
( interpolated pixel )の酉己置 につ レヽ て も 、 ί列 え ば'、 図 2 7 Α の よ う に 内 挿画素 Ρ 1 に 対 し て は 点線で 囲 ん だ領域 の 9 画素分 を 参 照 画素 と し て コ ン テ キ ス ト ( 二 0 〜 5 1 1 ) を 求 め 、 そ の コ ン テ キ ス ト に よ っ て 、 内 揷画素 P 1 を " 0 " か " 1 " か に決定す る 方法 も あ る 內揷画素 P 2 、 P 3 、 P 4 につ レ、 て も 、 そ れぞれの 内 挿位置 と そ の位置 を 囲 む直近 の 4 個 の 参 照 画素 の位 置 関係 に応 じ て 、 図 2 7 B 、 2 7 C お よ び 2 7 D の 点 線で 囲 ん だ画素 を 参照 画素 と す る 。
こ の 際 、 図 の よ う に 参 照 画素 A 〜 I の 、 内 揷画素 P に 対す る 相 対的 な位置 が 同 じ に な る よ う に配置 (例 え ば、 図 2 7 B は 図 2 7 Λ を 時計回 り に 9 0 度 回転 し た 配置 に な っ て レ、 る ) し て お け ば、 同 じ 画素ノ、。 タ ー ン を 回転 し た だ け の も の は 、 同 じ コ ンデ キ ス ト に な る の で 内挿画素 P 1 〜 P 4 で共通 の 内容 の テ ー ブル と し た メ モ リ を用 レヽ る こ と が で き る 。
次 に 、 4 X 4 画素 の マ ク ロ ブ ロ ッ ク を 1 6 X 1 6 画 素 に 拡大す る 場合 の例 を 説 明 す る 。
4 X 4 画素 の マ ク ロ ブ ロ ッ ク を 1 6 X 1 6 画素 に 拡 大す る に は 、 ま ず 、 4 X 4 の サ イ ズの マ ク ロ ブ ロ ッ ク M B を 8 X 8 の サ イ ズ に 拡大 し 、 次 に 8 X 8 の サ イ ズ の マ ク ロ ブ ロ ッ ク M B 力 ら 1 6 X 1 6 の サ イ ズの マ ク ロ ブ 口 ッ ク M B に 拡大す る と 云 う 手順 を踏む こ と に な る 。
こ の 4 X 4 の サ イ ズの マ ク ロ プ ロ ッ ク M B 力 ら 8 X 8 の サ イ ズ に 拡大す る 時 の 外部参 照 画素 (以 下 こ れ を border と 呼ぶ) A T と A L を 図 2 8 A に示 し 、 8 X 8 の サ イ ズの マ ク ロ ブ 口 ッ ク M B 力 ら 拡大す る 時 のタト 咅 15参照 画素 ( border ) B T と B L を 図 2 8 B に示す。
外部 参 照画素 A T は 2 行 8 列 、 外部参 照 画素
( border ) A L は 4 行 2 列 、 外部 参 照画素 B T は 2 行 1 2 列 、 外部参 照 画素 B L は 8 行 2 列 の 配列 で あ る 。
こ れ ら の外部参 照 画素 ( border )のィ直 は 図 2 9 を 用 い て 説 明 す る よ う に 、 既 に 再生 し た マ ク ロ プ ロ ッ ク の 所定の位置 の 画素 の 平均値 と し て 求 め る 必要 が あ る 。 し か し 、 2 値画像符 号化 回 路 1 7 0 と 2 値画像複 号化 回 路 2 8 0 に お い て 、 4 X 4 画素 の サ イ ズで符 号化す る 際 に は 、 外部参 照 画素 A T と A L を 参 照 し 、 8 X 8 画素 の サイ ズで符 号化す る 際 に は 、 外部参 照 画素 B T と B L を 参 照す る 場合 に は 、 こ れ を 転用 すれ ば拡大処 理の た め だ け に わ ざわ ざ外部参 照 画素 を 求 め る と 云 つ た 無駄 は 回避 で き る 。
こ こ で 、 外部参 照 B T と B L に つ い て は 、 平均値 を 求 め ず に 、 外部参照 画素 A T と A L 力ゝ ら 、 コ ン ビ ユ ー タ プ ロ ダ ラ ミ ン グ用 の 言語で あ る C 言語で記述すれ ば
B T [ 0 ] [ 0 ] = A T [ 0 ] [ 0 ]
Β Τ [ 0 ] [ 1 ] = Α Τ [ 0 ] [ 1 ]
Β Τ [ 1 ] [ 0 ] = Α Τ [ 1 ] [ 0 ]
Β Τ [ 1 ] [ 1 ] = Α Τ [ 1 ] [ 1 ]
Β Τ [ 0 ] [ 10 ] = A Τ [ 0 ] [ 6 ] ;
Β Τ [ 0 ] [ 11 ] = A Τ [ 0 ] [ 7 ] ;
Β Τ [ 1 ] [ 10 ] = A Τ [ 1 ] [ 6 ] ; B T [ 1 ] [ 1 1 ] = A T [ 1 ] [ 7 ] ;
o r ( j = 0 ; j < 2 ; j + + ) f o r ( i = 0 ; j < 8 + +
Figure imgf000095_0001
と い う 処理 に よ っ て 求 め る と 、 平均値 を 求 め る 場 合 と 多少結果 は異 な る が 、 平均値演算 を省 く こ と が で き る 上述 の 処理 は 、 図 2 8 A 、 2 8 B の 例 え ば 、 画素 p 1 を 画素 P 2 と 画素 p 3 に コ ピー し 、 画素 p 4 を 画素 P 5 と 画素 p 6 に コ ピ ーす る と レヽ う よ う に 、 A T や A L の値 を繰 り 返 し て 画素 を 追加処理 し て い uく こ と で 、 B T や B L を 生成す る も の で あ る 。 な お 、 C 言語プ ロ グ ラ ム 中 の " 门 [ ] " は配歹 IJ を 示 し 、 [ ] 中 の数字 は 1 0 進数 を示す。
以 上 は 、 拡大処理 の 具体的手法 を 説 明 し た 。 次 に 、 縮小処理 の 具体的 手法 を 説 明 す る 。
図 3 0 は 、 プ ロ ッ ク (マ ク ロ ブ ロ ッ ク ) を 縦横 " 1 Z 2 ,' の サイ ズ に縮小す る 縮小処理の 一例 で あ る 。 こ の 例 で は 、 点線枠 内 領域一 つ 一つ を 単位縮小 ブ ロ ッ ク 領域 と す る と 、 当 該 単位縮小 ブ ロ ッ ク 領域 内 の 2 X 2 画素 (各 点線枠 内 に お け る " O " で示 し た 計 4 画素) 毎 の 平均値 ( 図 3 0 の " X " ) を 、 そ の 単位縮小 ブ 口 ッ ク 領域で の 画素ィ直 と し て レヽ る 。 つ ま り 、 マ ク ロ プ ロ ッ ク を 縦横 " 1 4 " の サ イ ズ に縮小す る 場合 に は 、 単位縮小 プ ロ ッ ク 領域毎 に 、 そ の領域 内 の 4 X 4 画素 毎 の 平均値 を 求 め て 、 当 該単位縮小 ブ ロ ッ ク 領域 で の 画素値 と すれば よ い わ け で あ る 。
ま た 図 3 1 の よ う に 、 画素 A , B , C , D 力 ら な る , あ る 単位縮小 ブ ロ ッ ク 領域 を 考 え た場合 、 単位縮小 ブ 口 ッ ク 領域 の 画素 X の値 を 求め て こ の 単位縮小 プ ロ ッ ク 領域の値 と す る に あ た り 、 画素 A 〜 D の 平均値 を 求 め る の で は な く 、 こ れ よ り も 広 い範囲 の 画素位置 の 画 素 E , F , G , H , I , 】 , K , L , M , N , O , P を含 め て 、 こ れ ら A 〜 P の 平均値 を 求 め る よ う に し て も よ い。 つ ま り 、 隣接 の 単位縮小 ブ ロ ッ ク 領域 の 画素 の う ち 、 画素 A 〜 D に 隣接す る 画素 を含 め て こ れ ら の 平均値 を採用 す る 方式 と す る 。
以 上 は 、 現存す る 画素 の値の 平均値 を 単位縮小 プ ロ ッ ク 領域の 画素値 と す る こ と で 、 そ の 単位縮小 ブ ロ ッ ク 領域の 画素数 を 減 ら す よ う に し 、 マ ク ロ ブ ロ ッ ク の サイ ズ縮小 を 図 る よ う に し た例 で あ っ た。 こ の よ う な 、 計算 を 伴わず に 、 機械的 な 間 引 き ( thinning)処理 で も マ ク ロ ブ ロ ッ ク を縮小処理す る こ と 力 S で き る 。
そ の 例 を 次 に説 明 す る 。 は じ め に 、 マ ク ロ ブ ロ ッ ク 内 で 閉 じ た処理 を 説 明 す る 。
図 3 2 は 、 画素 間 引 き に よ る 縮小処理 の例 で あ る 。 1 6 X 1 6 画素構成 の マ ク ロ ブ ロ ッ ク M B は 画素 間 引 き に よ る 縮小処理 を 施 し た結果 、 8 X 8 画素構成 の ブ ロ ッ ク に縮小 さ れ た様子 を 示 し て い る 。 す な わ ち 、 図 中 で点線の 白 丸印 で示 さ れた 画素 が 間 引 かれた 画素 で あ り 、 実線の 白 丸印 で示 さ れた画素 が 単位縮小 ブ 口 ッ ク 領域の 画素値 ( 図 3 2 の " X " ) と な る 。 こ の 場合 縮小率 ( C R ) は 間 引 かれ る 画素 の割合 を 表す こ と に な る 。 な お 、 画素 間 引 き の 方法 は 図 3 2 で説 明 し た方 法 に 限 ら ず、 例 え ば、 サ イ コ ロ ( o n e of dice ) CO Έ. の 目 格子 ( 5 -spot f ace )状 に 画素 を 間 引 レ、 て も 良 レヽ 。 ま た 、 こ の 場合 、 拡大処理 と は 、 間 引 かれた 画素 を 内 挿す る こ と に相 当 す る 。
以 上 、 種 々 の縮小処理 の 具体的手法 を 説明 し た。 次 に拡大処理 に つ い て 説明 す る 。
図 3 3 は 、 拡大処理 を 表す図 で あ る 。 図 に お い て 、 実線方形枠 はマ ク ロ プ ロ ッ ク を示 し て お り 、 点線枠表 示 さ れた部分 の 升 目 一つ一 つ が 単位縮小 ブ ロ ッ ク 領域 を 示 し て い る 。 単位縮小 ブ 口 ッ ク 領域 内 の 白 丸 印 は 現 存画素 ( coded pixel )を 表 わ し て お り 、 内挿 に よ り 画素数 を増や し て マ ク ロ プ ロ ッ ク の 拡大 を 図 る よ う に す る 。 内 揷 さ れ る 画素 ( interpolated pixel )は
" X " で表 さ れて お り 、 內 挿 後 は " O " の 画素 は不 要 画素 と な る 。
マ ク ロ ブ ロ ッ ク 境界部 の 画素 を 内 揷す る 際 に 、 マ ク ロ ブ ロ ッ ク 外の 画素値が 必 要 に な る 。 こ の 場合 、 図 の 矢印 で示 さ れて レ、 る よ う に 、 マ ク ロ ブ ロ ッ ク 内 の 最 も 近 く の 画素値 を 割 り 当 て れば よ い。
す な わ ち 、 あ る 単位縮小 ブ ロ ッ ク 領域 の 画素 内 挿 を す る 場合 、 自 己 及び 自 己 と 隣接す る 各 単位縮小 プ π ッ ク 領域 で あ る 周 囲 8 プ ロ ッ ク の 画素値 の 計 9 画素 分 の 画素値力 s 、 処理 に 必要 で ¾> る 。 し 力 し 、 マ ク Π ブ 口 ッ ク 内 に お け る 単位縮小 ブ 口 ッ ク 領域が 、 マ ク π ブ π ッ ク 境界部 に位置 し て レ、 る と 、 周 囲 8 ブ ロ ッ ク の い く つ か は 自 己 の所属外の マ ク π ブ■ ッ ク 所属 の ブ ロ ッ ク と な る こ と か ら 、 自 己 所属 マ ク 口 ブ π ッ ク 外の 画素ィ直 を 別途 、 も ら つ て く る 必要 力 S あ る 。 こ の 場合 は 、 図 の 矢 印 で示 さ れて い る よ う に 、 自 己 の 所属 マ ク ロ ブ ロ ッ ク 内 の 画素 に最 も 近 く の 画素 の 画素値 を そ れぞれ周 辺 ブ ロ ッ ク に割 り 当 て て そ れ ら の 画素 直 を便宜的 に 画素値 の 処理 に 必要 な 近隣の 単位縮小 ブ 口 ッ ク 領域 で の 画素 値 と し利用 すれば よ い。
な お 、 マ ク ロ ブ ロ ッ ク 毎 の サ イ ズ変換処理 は 、 そ の マ ク ロ ブ Γ2 ッ ク 内 で 閉 じ て い る 必要 は な く 、 図 2 9 に 示 さ れ る よ う に 、 該 ブ 口 ッ ク 近傍 の 再生値 (左 · 上 - 左 上 ' 右上 に 隣接す る プ ロ ッ ク 内 の 画素値) を用 レヽ る よ う に し て も 良 い o
こ の こ と を 具体的 に説明 す る 。
図 2 9 に お い て 、 実線方形枠 は あ る ブ ロ ッ ク ( マ ク π プ ロ ッ ク ) を 示 し 、 " X " 印 は倍率 1 倍の と き の 画 像 (標 準倍率画像) の 各画素 を 示 し て い る 。 マ ク ロ ブ ロ ッ ク は通 常 、 1 6 X 1 6 画 素構成 で あ り 、 フ レ ー ム を 1 ノ 2 に圧縮 し し 口 に ブ ロ ッ ク 内 の 画素 は 8 X 8 画素構成 に な り 、 先 の 1 6 X 1 6 画素構成 は そ れぞれ 9
2 X 2 画素分 の 画像 が 1 画素 で表現 さ れ る こ と に な る の で 、 こ の場合 の 2 X 2 画素分 の 画像 を 代表 点 の 情報 を 1 画素表現形式 に し て 表 し た の が 図 2 9 に お け る Ο " 印 で あ る 。 点線で示 し た枠 内 が 、 単位縮小 ブ 口 ソ ク 領域で あ っ て 、 標 準倍率画像 で の 4 画素構成 ( 2 X 2 ) の領域 を示 し て お り 、 1 / 2 縮小 の 場合 は こ の 点 線枠 内領域毎 に 1 画素 で表現 さ れ る 。
1 / 2 縮小画像 を 元 の 画像サ イ ズ に復元す る 場合 標準倍率画像 に復元す る 場合) 、 1 / 2 縮小画像 の 画素 の領域 は 4 画素構成 に 戻す が 、 そ れ を マ ク ロ ブ ッ ク 内 で 閉 じず に 内 揷 に よ り 行 う に は次 の よ う にす る
例 え ば、 図 2 9 に お い て 、 あ る マ ク ロ ブ ロ ッ ク の 、 あ る 単位縮小 ブ ロ ッ ク 領域 (点線枠 内領域) の 画素 1 や画素 2 を 内 揷 に よ り 再生す る 場合 を 考 え て み る 。 こ の 場合 、 そ の 処理時 点 で現存す る の は " o " 印 の 画素 小±報で あ る 。 従 っ て 、 内揷 し ょ う と す る 画素 1 や画素
2 の周 囲 に あ る 現存画素 ( c o d e d p i X e 1 )は画素 3 , 4 , 5 お よ び 6 で あ る 力ゝ ら 、 こ れ ら 画素 3 , 4 , 5 お ぶ び 6 を用 い て 双 一 次 内 挿 を 行 う 。 た だ し 、 "画素 3 " や "画素 4 " は 隣接 ブ ロ ッ ク ( 隣接す る マ ク ロ ブ α ッ ク ) に所属す る 画素 で 、 し か も 、 拡大前 の 画素
( 1 / 2 縮小画像 の 画素) で あ り 、 隣接 プ ロ ッ ク は 時 間 的 に 前の 時点 で処理 さ れ る 位置 に あ る プ ロ ッ ク で あ る の で 、 こ の マ ク ロ ブ ロ ッ ク の 拡大処理後 に は 、 不要 と な る デー タ で あ る 力 ら 、 メ モ リ リ ソ ー ス の 節約 の観 点 力 ら 、 既 に廃棄 さ れて し ま っ て レ、 る こ と 力 S あ る 。
そ の よ う な シ ス テ ム の 場合 に は 、 例 え ば、 "画素
3 " で 云 え ばそ の マ ク ロ ブ ロ ッ ク で の 内挿済 み で現存 す る 隣接画素 で あ る 画素 7 , 8 , 9 お よ び 1 0 の 平均 値 を 求 め て 、 前記廃棄 し て し ま っ た "画素 3 " の値 と す る の も 一つ の 方法 で あ る が 、 そ の 平均値 を 求 め る 演 算 を少 な く し た レヽ場合 に は 、 画素 7 , 8 , 9 お よ び 1 0 の 計 4 画素 の う ち 、 内 挿 し ょ う と す る 画素 1 , 2 に 近い 、 画素 9 お よ び 1 0 の 平均値 を 、 廃棄 し て し ま つ た "画素 3 " の値 と し て 用 レ、 る よ う に し て も よ い。
ま た 、 画素 1 0 を そ の ま ま 流用 し て "画素 3 " の値 と す る よ う にす る と 、 さ ら に演算 を省 け る 。 "画素
4 " に つ い て も 、 同様 に 、 そ の 近 く の 画素値 で代用 す る 。 ま た 、 同様 な ケ ー ス で 、 "画素 1 1 ,, の 内 挿 に は 例 え ば、 "画素 1 2 " の代わ り に 画素 1 4 と 画素 1 5 の 平均値、 "画素 1 3 " の代わ り に 画素 1 6 、 画素 1 8 の代 わ り に 画素 1 7 を 用 い る 。
フ レ ー ム 画像 P f は通 常 、 オブ ジ ェ ク ト 部分 を 主体 と し た最小 の矩形範囲 、 例 え ば、 図 3 4 A に示 さ れ る 符 号化領域 (video obj ect plane ) C A 内 を ブ ロ ッ ク (マ ク ロ ブ ロ ッ ク ) に 分割 し て 符 号化す る た め 、 符 号化領域の 境界部 に位置す る プ ロ ッ ク で は 、 該 ブ 口 ッ ク の左 ' 上 ' 左 上 ' 右上 に 隣接す る ブ ロ ッ ク が 、 符 号 ィ匕領域 C A の外部 に位置す る こ と と な っ て し ま う 場合 が あ る 。
こ の 場合 、 図 2 9 の よ う に 、 ブ ロ ッ ク 近傍の 再生画 素値 (再生 し た画素 の値) を利用 す る 場合 で も 、 符 号 ィ匕領域 C A の外部 に位置す る マ ク ロ ブ 口 ッ ク で あ る 場 合 に は再生両素値 を 参 照せず に 、 図 3 3 の よ う に 自 己 所属 マ ク ロ ブ 口 ッ ク 内 の 画素 に最 も 近 く の 画素 のィ直 を 便宜的 に そ の プ ロ ッ ク に割 り 当 て て 用 レヽれば よ レヽ 。
更 に 、 誤 り の影響 を 受 け る 伝送路 でデー タ の授受 を 行 う 場合 に は 、 誤 り の影響 を 受 け難 く す る た め に 、 符 号化領域 C A よ り も 小 さ い 単位 ( こ れ を " 同 期 回復 単 位 (video packet ) " と 呼ぶ) で符 号化処理 を 閉 じ る 場合が あ る 。
こ れ に よ り 、 誤 り の影響 を こ の " 同 期 回復 単位 " で 断 ち 切 る こ と が 可能 と な り 、 誤 り の影響 を 受 け難 く な る 。 こ こ で 、 " 同 期 回復 単位 " と は 、 図 3 4 B に U n な る 符 号 を 付 し て 示 し た 点線で 囲 ま れ る 各領域 を指す " 同期 回復 単位 " は符 号化領域 C A を 小 さ く 区切 つ た 領域で あ る が 、 複数 の マ ク ロ ブ 口 ッ ク で構成 さ れ る こ と は変 わ り な い。
こ の 方式の場合 、 " 同 期 回復 単位 " 内 で符 号化処理 を 閉 じ る た め 、 あ る " 同 期 回 復 単位 " で使用 す る デー タ に伝送誤 り が あ る も の が含 ま れて レヽ た と し て も 、 そ の誤 り の あ る デー タ を 参照 し て 処理す る の は " 同 期 回 復単位 " 内 だ け で あ り 、 隣接 の " 同 期 回復 単位 " は誤 り の あ る デー タ を 参 照 し て 処理す る こ と は な い 力 ら 、 伝送誤 り が波及 し に く い 処理方式 と な る 。
こ の 場合 で も 、 図 2 9 の よ う に 、 該 プ ロ ッ ク 近傍 の 再生画素値 を利用 す る 際 に 、 該 ブ ロ ッ ク が含 ま れ る " 同 期 回復単位 " 以外の 同 期 回復 単位 に 属す る マ ク ロ ブ ロ ッ ク の 再生画素値 を 参 照せず に 、 図 3 3 の様 に マ ク ロ ブ 口 ッ ク 内 の 最 も 近 く の 画素値 を割 り 当 て る 。
上述 し た よ う な 、 "マ ク ロ ブ ロ ッ ク や 同 期 回復 単位 の 外側 を 参照す る か 、 し な レ、 か " は 、 符 号 に 切 り 換 え ビ ッ ト を 用 意 し 、 こ れ を 用 レ、 て 切 り 換 え る よ う にす る と 伝送エ ラ ー の頻度や 、 許容 さ れ る 演算 量 、 メ モ リ 量 な ど の様 々 な 状況 に 対応す る こ と が で き る 。
と こ ろ で 、 先 に述べた よ う に双一次 内揷 で は 、 周 囲 の 4 画素 の み を用 い て 処理 を行 っ て レ、 る 。 そ の た め 、 再生 さ れた画素 に よ る 画像 に は 、 特 に斜 め 方 向 の 不連 続性が発 生 し 、 視覚 的劣化 が 起 き 易 い傾 向 が避 け ら れ な い。 こ れ を 回避す る に は例 え ば 、 図 3 3 の 内 挿対象 画素 を 内挿す る 例 を 取 り 上 げて み る と 、 そ の 内 挿 の 際 に 、 双 一 次 内挿 の 参 照範囲 よ り も 広 い 拡大参 照範囲 に 含 ま れ る 画素 を用 い て 内 挿す る 。
す な わ ち 、 双一 次 内挿 ( bi-linear
interpolation )の 参 照範囲 よ り も 広 い 拡大参照範囲 に含 ま れ る 画素 を 用 い て 内 挿す る こ と で 、 よ り 広 い範 囲 の 画素 の傾 向 が反 映 さ れ る こ と に な り 、 不連続性 の 問題 が 回避 さ れ る 。 ま た 、 内 挿 に用 い る 画素 の 数 が " 4 画素 " の よ う に 偶数個 で あ る よ り も 、 " 9 画素 " の よ う に奇数個 の方が 、 多数決効果 が 得 ら れ 、 不連続 性 の 回避効果が よ り 一層 顕著 に得 ら れ る 場合 力 s あ る 。
図 2 6 B に 、 1 2 個 の 画素 を 用 い て 内 挿 を 行 う 例 を 示す。 先 に 図 2 6 A を 用 い て 説明 し た 実施例 の 表記 を 用 い る と 、 あ る マ ク ロ ブ ロ ッ ク の 1 / 2 縮小画像 に お け る マ ク ロ ブ 口 ッ ク 内 あ る 位置 で の 再生画素 を P 1 , p 2 , p 3 , ρ 4 、 そ し て 、 そ れ ら の 直 (画素値) を I pi, I p2, I p3, I p4 と し た 場合 に 、 こ れ ら I 1 , I p 2 , I p 3 , I p4 は 、
pi : if ( 4 * I a + 2 * ( I b + I c + I d + I e + I f + I g + I h + I i + I + I k + 1 1 ) > 8 then " 1 " else " 0 "
p2 : if ( 4 * I b + 2 * ( I a + I c + I d + I e + I f + I g + I h + I i + I + I k + 1 1 ) > 8 then " 1 " else " 0 "
p3 : if ( 4 * I c + 2 * ( I b + I a + I d + I e + I f + I g + I h + I i + I j + I k + 1 1 ) > 8 then " 1 " else " 0 "
p4 : if ( 4 * I d + 2 * ( I b + I c + I a + I e + I f + I g + I h + I i + I j + I k + I 1 ) > 8 then " 1 " else " 0 "
で表せ る 。 た だ し 、 a は画素 A 、 b は画素 B 、 c は 画 素 C 、 d は画素 D 、 e は画素 E 、 f は画素 F 、 g は画 素 G 、 h は画素 H 、 i は画素 I 、 j は画素 J 、 k は画 素 K 、 1 は画素 L を 示す。 0 2 ま た 、 " 1 / 4 " サ イ ズの サ イ ズ変換処理 は 、 " 1 / 2 " サ イ ズ の サイ ズ変換処理 を 2 回施す こ と で実現 し て も 良 い
次 に 、 フ レ ー ム 単位 の サ ィ ズ変換処理 と の組み合 わ せ の 例 を 、 第 5 の 実施例 と し て 説明 す る 。
本発 明者 ら が 特願 平 8 — 2 3 7 0 5 3 号で 開 示 し た 技術 に は 、 フ レ 一 ム 単位 (実際 に はォブ ジ ェ ク ト を含 む方形領域 ) でサ ィ ズ変換 を 施す こ と で レ 一 ト 制御す る 実施例 と 、 プ 口 ッ ク の よ う な 小領域単位 で サ イ ズ変 換 を 施す こ と で レ 一 ト 制御す る 実施例 と を 提示 し て い る 。 ま た 、 上述 の 第 1 乃 至第 3 の 実施例 に は 、 さ ら に 具体的 な 小領域単位 で サイ ズ変換 を 施す こ と に よ り 、 レ ー ト 制御す る よ う に し た例 を示 し た。
こ こ に説明す る 本実施例 ^し レ 1 て は 、 フ レ ー ム 単位 で の サ イ ズ変換処理 と 、 小領域単位 で の サ ィ ズ変換処 理 を組み合 わせ て使用 す る 例 を 示す も の で あ る 。
図 3 5 は 、 本実施例 の ァ ル フ ァ マ ッ プ符 号化装置 を 説明 す る 図 で あ る 。 こ の装置 は縮小 回 路 5 3 0 , 5 2 1 , 5 2 6 、 2 値画像符 号化 回 路 5 2 2 、 拡大 回 路 5 2 3 , 5 4 0 、 動 き 補償予測 回 路 5 2 5 、 多重化 回 路
5 2 4 , 5 5 0 と よ り 構成 さ れ る
こ の よ う な構成 に お い て 、 ァ ル フ ァ マ ッ プ信 号入力 線 2 を 介 し て 供給 さ れ る 2 値化 さ れた ァ ノレ フ ァ マ ッ プ 画像 は 、 縮小 回 路 5 3 0 に お レ、 て サ イ ズ変換率 C R に 基づ レ、 て フ レ 一 ム 単位 に縮小 さ れ る 。 フ レ ー ム 単位 に 縮小 さ れた信 号 は 、 信 号線 5 0 2 を 介 し て ア ル フ ァ マ ッ プ符 号化 回 路 5 2 0 に供給 さ れ、 小領域 に 分割 さ れ た後 、 符 号化 さ れ る 。
こ こ で 、 ア ル フ ァ マ ッ プ符 号化 回 路 5 2 0 は 、 図 1 2 の ア ル フ ァ マ ッ プ符 号化 回 路 2 0 と 等価 で あ り 、 ァ ル フ ァ マ ッ プ符 号化 回 路 5 2 0 の構成要素 5 2 1 〜 5 2 6 は 、 そ れぞれ図 1 2 の ア ル フ ァ マ ッ プ符 号化 回 路 2 0 の構成要素 2 1 〜 2 6 と 同 じ機能 を 有す る の で 、 ア ル フ ァ マ ッ プ符 号化 回 路 5 2 0 の 説 明 は こ こ で は省 略す る 。 ま た 、 図 1 2 の ア ル フ ァ マ ッ プ符 号化 回 路 2 0 の構成 は 、 図 1 の ア ル フ ァ マ ッ プ符 号化 回 路 2 0 の 構成 を簡 略化 し て 表現 し た も の で あ る た め 、 ア ル フ ァ マ ッ プ符 号化 回 路 5 2 0 の構成 は 、 図 1 の ア ル フ ァ マ ッ プ符 号化 回 路 2 0 と 等価 な構成 で も 良 い。
ア ル フ ァ マ ッ プ符 号化 回 路 5 2 0 で符 号化 さ れた符 号化情報 に は 、 小領域毎 の サ イ ズ変換率 C R b が 多重 ィ匕 さ れて お り 、 信 号線 5 0 3 を 介 し て 多重化 回 路 5 5 0 に供給 さ れ、 フ レ ー ム 単位 の サ イ ズ変換率 C R の 符 号化情報 と 多重化 さ れ、 信 号線 3 を介 し て 出 力 さ れ る 。
ま た 、 ア ル フ ァ マ ッ プ符 号化 回 路 5 2 0 の 再生画像 は信 号線 5 0 4 を 介 し て 拡大 回 路 5 4 0 に供給 さ れ、 フ レ ー ム 単位 の サ イ ズ変換率 C R に 基づ い て 拡大 さ れ た後 、 信 号線 4 を 介 し て 出 力 さ れ る 。
図 3 6 は 、 本実施例 の 復 号化装置 を 説 明 す る 図 で あ る 。 こ の ア ル フ ァ マ ッ プ複 号化装置 は 、 分離化 回 路 6 5 0 , 6 4 3 、 2 値画像復 号化 回 路 6 4 1 、 縮小 回 路
6 4 5 、 動 き 補償予測 回 路 6 4 4 、 拡大 回 路 6 4 2 , 6 6 0 と 力 ら 構成 さ れて い る 。
こ の よ う な構成 に お い て 、 信 号線 8 を介 し て 供給 さ れ る 符 号化情報 は 、 分離化 回 路 6 5 0 で フ レ ー ム 単位 の サ イ ズ変換率 C R と 、 小領域単位 の符 号化情報 に分 離 さ れ る 。 小領域単位 の 符 号化情報 は 、 信 号線 6 0 8 を 介 し て ア ル フ ァ マ ッ プ復 号化 回 路 6 4 0 に供給 さ れ 、 小領域単位 の 再生信号が 信 号線 6 0 9 を 介 し て 拡大 回 路 6 6 0 に供給 さ れ る 。
ア ル フ ァ マ ッ プ複 号化 回 路 6 4 0 は 、 図 1 3 の ア ル フ ァ マ ッ プ復 号化 回 路 4 0 と 等価 な た め 、 こ こ で は説 明 を省略す る 。 ま た 、 図 1 3 の ア ル フ ァ マ ッ プ複 号化 回路 4 0 の構成 は 、 図 2 の ア ル フ ァ マ ッ プ復 号化 回 路 4 0 の構成 を簡略化 し て 表現 し た も の で あ る た め 、 ァ ル フ ァ マ ッ プ復号化 回 路 6 4 0 の構成 は 、 図 2 の ア ル フ ァ マ ッ プ複号化 回 路 4 0 と 等価 な構成 で も 良 い。
拡大 回 路 6 6 0 で は 、 フ レー ム 単位 の縮小率 C R の 情報 に 基づ い て 、 信 号線 6 0 9 を 介 し て 供給 さ れ る 再 生信号 を 拡大処理 し て 、 信 号線 9 よ り 出 力 さ せ る 。
こ の よ う に 、 こ こ で は ア ル フ ァ マ ッ プ信 号 を フ レ ー ム 単位 でサ イ ズ変換処理 し 、 かつ 、 小領域単位 でサ イ ズ変換処理す る よ う に し た。 こ の フ レー ム 単位 で の サ ィ ズ変換 と 、 小領域単位 で の サ イ ズ変換 と を組み合わ せ る こ と で 、 サ イ ズ変換率情報 の よ う な サ イ ド情報 が 削減 さ れ る た め 、 特 に 、 低 い符 号化 レ ー ト で符 号化す る 場合 に効果的 と な る
次 に フ レ— ム メ モ リ につ い て 、 触れて お く 。
図 3 5 お よ び図 3 6 に は 明 示 し て い な い が 、 符 号化 装置 と 復号化装置 に は いずれ も 再生画像 を 蓄積す る た め の フ レ ー ム メ モ リ が 必要 で あ る 。 図 3 7 は 、 フ レ 一 ム 毎 の解像度 の 例 を 表 し た 図 で あ る。 本発 明 で は 、 動 き 補償予測 を 用 い て い る た め 、 例 え ば、 時刻 n の フ レ ー ム を符 号化す る 際 に は 、 時刻 n 1 の フ レ ー ム の解 像度 を 時刻 n の フ レ一ム で の 解像度 ( こ の場合 、 サ ィ ズ変換率) に一 致 さ せ な け ればな ら な い で 、 フ レ 一 ム メ モ リ に 蓄積す る 再生画像が 、 ァ ル フ ァ マ ッ プ 符 号化 回 路 5 2 0 に よ つ て 図 3 7 に示 さ れ る よ う に フ レ ー ム 単位の解像度 ( こ の 例 の 場合、 時刻 n の フ レ 一 ム な ら ば C R = 1 / 2 ( 図 3 7 の ( a ) ) 、 時刻 n の フ レ ー ム な ら ば C R = 1 ( 図 3 7 の ( c ) ) で 蓄積 さ れ る 場合 と 、 元 の解像度 (時刻 に 関 わ ら ず 、 常 に サ イ ズ変換率 C R = 1 ) で さ れ る 場合 の 2 通 り が 考 え ら れ る 。
前者 に お い て は 、 フ レ ー ム メ モ リ は信 号線 5 0 4 お よ び信 号線 6 0 9 を 介 し て 供給 さ れ る フ レー ム 単位 の 解像度 の 再生画像 を 蓄積す る こ と に な り 、 後者 に お け る フ レ ー ム メ モ リ は 、 信号線 4 お よ び信 号線 9 を 介 し て 供給 さ れ る 元 の解像度 の 再生画像 を 蓄積す る こ と に な る 。 従 っ て 、 図 3 5 の 符 号化装置 お よ び図 3 6 の 復 号化 装置 に お い て フ レ ー ム メ モ リ を 明 示す る と 、 前者の フ レ ー ム メ モ リ ( こ れ を F M 1 (符 号化装置用 ) 、 F M 3 (復 号化装置用 ) と 表す こ と と す る ) の場合 は 、 図 3 8 お よ び図 3 9 の よ う に な り 、 後者 の フ レ ー ム メ モ リ ( こ れ を F M 2 (符 号化装置用 ) 、 F M 4 (復 号化 装置用 ) と 表す こ と と す る ) の場合 は 、 図 4 0 お よ び 図 4 1 の よ う に な る 。
つ ま り 、 図 3 8 の符 号化装置 は 、 サ イ ズ変換率 C R の 情報 と 拡大 回路 5 2 3 の 出 力 情報 を F M 1 な る フ レ ー ム メ モ リ に保持 し 、 こ の保持情報 を M C (動 き 補償 予測 回 路) 5 2 5 に 与 え る よ う に構成 さ れ、 図 3 9 の 復 号化装置 は 、 C R の 情報 と 拡大 回 路 6 4 2 力ゝ ら の 出 力 を F M 3 な る フ レ ー ム メ モ リ に保存 し 、 こ の保存 出 力 を 動 き 補償予測 回 路 6 4 4 に 与 え る よ う に構成 さ れ て レヽ る 。
ま た 、 図 4 0 の符 号化装置 は 、 サ イ ズ変換率 C R の 情報 と 拡大 回 路 5 4 0 の 出 力 情報 を F M 2 な る フ レ ー ム メ モ リ に保持 し 、 こ の保持情報 を M C (動 き 補償予 測 回路) 5 2 5 に 与 え る よ う に構成 さ れ、 図 4 1 の 復 号化装置 は 、 C R の 情報 と 拡大 回 路 6 6 0 か ら の 出 力 を保存 し 、 こ の保存 出 力 を動 き 補償予測 回 路 6 4 4 に 与 え る よ う に構成 さ れて い る 。
フ レ ー ム メ モ リ F M 1 , F M 3 の 具体的 な 構成 は 、 図 4 2 A に示 さ れ、 フ レ ー ム メ モ リ F M 2 , F M 4 の 具体的 な構成 は 、 図 4 2 B に示 さ れて レ、 る 。
フ レ ー ム メ モ リ F M 1 , F M 3 は 、 図 4 2 A に示す よ う に 、 現 フ レ ー ム の 画像 を保存す る フ レ ー ム メ モ リ m i l と 、 こ の フ レ ー ム メ モ リ m i l の保持画像 を別 途与 ら れ る サ イ ズ変換率 ( size conversion ratio ) C R の 情報対応 に サ イ ズ変換処理 ( s i z e conversion process )す る ケ づ ス 変換回 路 ( size conversion circuit ) m 1 2 、 こ の サ イ ズ変換回 路 m 1 2 でサ イ ズ変換処理 さ れた 出 力 を保存す る フ レ ー ム メ モ リ m l 3 と 力 ら な り 、 ま た 、 フ レ ー ム メ モ リ F
M 2 , F M 4 は 、 図 4 2 B に示す よ う に 、 現 フ レ ー ム の 画像 を保存す る フ レ ー ム メ モ リ m 2 1 と 、 こ の フ レ ー ム メ モ リ m 2 1 の保持画像 を 、 別途与 え ら れ る サ イ ズ変換率 C R の 情報対応 に縮小処理す る 縮小 回 路 m 2 2 、 こ の縮小回 路 m 2 2 で縮小処理 さ れた 出 力 を保存 す る フ レ ー ム メ モ リ m l 2 と 力 ら な る 。
こ の よ う な構成 の フ レ ー ム メ モ リ に お け る 動作 を 説 明 す る 。
ま ず フ レ ー ム メ モ リ F M 1 , F M 3 の 場合 、 こ れ ら に は信 号線 5 0 4 お よ び信号線 6 0 9 を 介 し て フ レ ー ム 単位 で そ の フ レ ー ム の 対応す る 解像度 の 再生画像 が 供給 さ れ、 現 フ レ ー ム 保存用 の フ レ ー ム メ モ リ m l 1 に 蓄積 さ れ る 。 フ レ ー ム メ モ リ m l 1 は 、 現 フ レ ー ム の 符 号化 (例 え ば 、 時刻 n ) が 終 了 し た 時点 で現 フ レ — ム の 再生画像 が 全て 蓄積 さ れて レ、 る 。 0 8 次 に 、 サ ィ ズ変換回 路 m 1 2 で は 、 時刻 n + 1 に お け る フ レ一ム の 符 号化 が 開 始 さ れ る 時点 で、 時刻 n に お け る フ レ — ム の 再生画像 を フ レ ー ム メ モ リ m l 1 力 ら SJCみ込 み 、 時刻 n + 1 に お け る フ レ ー ム の フ レ ー ム 単位で の サィ ズ変換率 C R と 一致す る よ う に 、 サ イ ズ 変換処理 (解像度変換) を 行 う
図 3 7 A〜 3 7 D の 例 に お い て 、 図 3 7 A に示す よ う に 時刻 n に お け る フ レ ー ム の サ イ ズ変換率 C R は
2 " で あ り 、 ま た 、 図 3 7 B に示す よ う に そ の 次 フ レ一ム で あ る 時刻 n + 1 に お け る フ レ ー ム の サ イ ズ変換率 C R は " 1 " で あ る の で 、 こ の 場合 、 サ イ ズ 変換回 路 m 1 2 で はサイ ズ変換率 C R を " 1 Z 2 " か ら " 1 " に変換す る 処理 を行 う こ と に な る 。
サ ィ ズ変換回路 m 1 2 で解像度変換 さ れた 時刻 n に お け る 再生画像 は 、 前 フ レ ー ム の フ レ ー ム メ モ リ m l 3 に 蓄積 さ れ、 時刻 n + 1 に お け る フ レ ー ム の 動 き 補 償予測 の 参照 画像 と な る
フ レ ― ム メ モ リ F M 2 , F M 4 の場合 、 こ れ ら に は 信号線 4 お よ び信号線 9 を 介 し て 元 の解像度 ( C R = 1 ) の 再生画像 が供給 さ れ、 現 フ レ ー ム の フ レ ー ム メ モ リ m 2 1 に 蓄積 さ れ る 。 フ レ ー ム メ モ リ m 2 1 に は 現 フ レ一ム の符号化 (例 え ば 、 時刻 n ) が 終 了 し た 時 点 で現 フ レ ー ム の 再生画像 が 全て 蓄積 さ れて レ、 る 。
次 に 、 縮小 回路 m 2 2 で は 、 時刻 n + 1 に お け る フ レ 一 ム の 符 号化 が 開 始 さ れ る 時点 で、 時刻 n に お け る 0 9 フ レ 一 ム の 再生画像 を フ レ ー ム メ モ リ m 2 1 か ら 銃 み 込 み 、 時刻 n + 1 に お け る フ レ ー ム の フ レ ー ム 単位 で の サ イ ズ変換率 C R と 一致す る よ う に 、 縮小処理 (解 像度変換) を行 う
フ レ ー ム メ モ リ m 2 1 に 蓄積 さ れ る 再生画像 は 常 に サ イ ズ変換率 C R = 1 で あ る の で、 図 3 7 A〜 3 7 D の例 で は 、 時刻 n + 1 に お け る サ イ ズ変換率 C R は で あ る カゝ ら 、 こ の 、 縮小 回 路 m 2 2 で は解 像度変換処理 を 行 わ な い 。 な お 、 図 3 7 A〜 3 7 D の 例 の場合 、 現 フ レ一ム が 時刻 η + 1 の 、 時刻 n + 2 に お け る フ レ ー ム の 解像度 は " 1 / 2 " で あ る カゝ ら 縮小 回 路 m 2 2 で はサ イ ズ変換率 C R を " 1 " ら
2 " にす る 解像度変換処理 を 行 う こ と に な る 。 縮小 回路 m 2 2 で 解像度変換 さ れた 時刻 n に お け る フ レ ー ム の 再生画像 は 、 前 フ レ — ム の フ レ ー ム メ モ リ m 3 3 に 蓄積 さ れ、 時刻 η + 1 に お け る フ レ一ム の 動 き 捕償予 測 の 参照画像 と な る 。
以 上、 フ レ ー ム メ モ リ F Μ 1 , F M 2 , F M 3 , F M 4 の 具体的 な構成 と 作用 を 述ベ た カ 、 フ レ ー ム メ モ リ F M 1 , F M 3 は、 信 号線 5 0 4 お よ び信 号線 6 0 9 を 介 し て 供給 さ れ る フ レ ー ム 単位 の解像度 の 再生画 像 が 蓄積 さ れ る こ と が 特徴で あ り 、 ま た 、 フ レ ー ム メ モ リ F M 2 , F M 4 は 、 信 号線 4 お よ び信 号線 9 を 介 し て 供給 さ れ る フ レ ー ム 単位 で の解像度 の 再生画像 が 蓄積 さ れ る こ と 力 S 特徴 で あ る た め 、 そ れぞれ、 他 に 種 々 の構成 が 考 え ら れ る 。
次 に 、 マ ク ロ ブ 口 ッ ク の 属性情報 ( m o d e
inf ormation )を 符 号化す る 方式の例 を 第 6 の 実施例 と し て 説明 す る 。 ま ず 、 特願平 8 — 2 3 7 0 5 3 号 に て 、 提案 し た方式の 説 明 を す る 。
図 4 3 A お よ び 4 3 B は 、 時刻 n と 時刻 n — 1 に お け る あ る マ ク ロ ブ 口 ッ ク の 属性情報 の 一例 を 表 し て い る 。 伹 し 、 こ こ で 云 う 属性情報 と は 、
" transparent" ( そ の マ ク 口 ブ ロ ッ ク の構成画素 全て が透明 ) 、 " opaque" ( そ の マ ク ロ ブ ロ ッ ク の 構成画素全て が 不透明 ) 、 " Mu 11 i " ( そ の マ ク ロ ブ 口 ッ ク の構成画素 は透 明 · 不透明 混在) と い っ た マ ク ロ ブ 口 ッ ク の 中 味 の 状態 を 示 し た情報 で あ る 。
例 え ば、 " transparent" を符 号 " 0 " で 、
" o aque" を符 号 " 3 " で 、 そ し て 、 " M u 1 t i " を 符 号 " 1 " で ラ ベル付 け し て 示 し た例 で あ る 。
フ レ ー ム に お け る オ ブ ジ ェ ク ト 部分 を含 む最小矩形 ( tightest rectangle area ) 〖こ 目 し 、 そ の 左 上 を 、 領域の境界部 に接す る よ う に 、 方形領域 を 設 定す る と 、 当 該設 定 し た方形領域 に含 ま れ る 各 マ ク 口 ブ ロ ッ ク の 属性情報 の 分布 ( ラ ベルの 分布) は 、 例 え ば、 図 4 3 A お よ び 4 3 B の よ う に な る 。
そ し て 、 図 4 3 A に示 し た 時刻 n の フ レ ー ム で の マ ク ロ ブ ロ ッ ク の構成画素属性情報の 分布例 と 、 図 4 3 B に示 し た 時亥 lj n — 1 の フ レ ー ム で の マ ク ロ プ ロ ッ ク の構成画素属性情報 の 分布例 の よ う に 、 時 間 的 に 近い フ レ一ム の ァ ノレ フ ァ マ ッ プ間 で は 、 非 常 に似 た ラ ベル 付 け が行わ れ る 。
従 つ て 、 こ の よ う な場合 に は 、 フ レ ー ム 間 に お レヽ て ラ ベル の 相 関 が 高 い た め 、 既 に符 号化済 みの フ レ ー ム の ラ ベ ノレ を禾 ϋ用 し て 、 現 フ レ ー ム の ラ ベ ル を 符 号化す る と で 、 大幅 に符 号化効率 が 改善 さ れ る こ と に な る ま た 、 一般 に 、 時刻 n の フ レ ー ム に お け る 符 号化領 域 ( ォプ ジ ク ト 部分 を 主体 と し た最小 の 矩形範囲 、 例 え ば、 図 3 4 A に示 さ れ る 符 号化領域 C A ) と 、 時 刻 n の フ レ ー ム に お け る 符 号化領域 の サ イ ズが 異 な る ·¾ 口 力 S ぁ る 。 こ の 場合 、 一例 と し て 、 図 4 4 A お よ び 4 4 B に示す手順 で 、 時刻 n — 1 の フ レ ー ム で の 符 号化領域サ イ ズ を 時刻 n の フ レ ー ム で の サ イ ズ に合 わせ る 。 例 え ば、 時刻 n の フ レ ー ム に お け る 符 号化領 域サ ィ ズが 、 時刻 n ― 1 の フ レ ー ム に お け る 符 号ィ匕領 域サ ィ ズ に比べ て行 が 1 行分長 く 、 列 が 1 列分短い場 合 は 、 図 4 4 A の よ う に 、 行 の短い 時刻 n — 1 の フ レ ― ム に け る 符 号化領域 に つ い て 、 そ の領域 内 右端 の
1 列分の マ ク ロ ブ ロ ッ ク 列 を カ ッ ト し 、 そ の 後 、 下部 の 1 行分 の マ ク ロ プ ロ ッ ク 列 を そ の 下 に コ ピー し て 行 を増やす 。 こ の 状態 が 図 4 5 B に示 さ れて い る 。
ま た 、 時刻 n — 1 の フ レ ー ム に お け る 符 号化領域サ ィ ズが 、 時刻 n の フ レ ー ム に お け る 符 号化領域サ イ ズ よ り 列 につ い て は 1 列分短 く 、 行 に つ い て は 1 行分長 い場合 は 、 符 号化領域 内 の 下端の 1 行分 の マ ク ロ ブ 口 ッ ク 列 を カ ッ ト し 、 そ の 後 、 そ の 符 号化領域 内 の右端 の 1 列分 の マ ク ロ ブ 口 ッ ク 列 を そ の 隣 り に コ ピ ー し て 1 列増やす。
サイ ズが 合わ な い と き は 、 こ の よ う に し て サ イ ズ を 合 わせ る 。 な お 、 サ イ ズ の 合 わせ方 は前記の 方 法 に 限 つ た も の で は な い。 最終的 に 、 図 4 4 B の様 に 、 時刻 n の フ レ ー ム で の サ イ ズ に 合 わせ ら れた 、 時刻 n — 1 の フ レ ー ム で の ラ ベ ル を 、 こ こ で は便 宜上 、 時亥 U n — 1 ' の ラ ベル と 表記 し て 以 下の 説 明 に用 い る こ と にす る 。
図 4 6 A は 、 時刻 n で の 上述 の マ ク ロ プ ロ ッ ク の 属 性情報 と 、 時刻 n _ 1 ' で の 上述 の マ ク ロ ブ ロ ッ ク の 属性情報 の 差分 、 つ ま り 各 マ ク 口 プ ロ ッ ク 位置 で の 各 ラ ベ ル の 差分 を 、 同 一位置 の も の 同 士 で と っ た結果 を 示 し て い る 。
こ こ で 、 " S " は " ラ ベル力 S —致 し て い る " こ と を 示 し 、 " D " は " ラ ベルが 不 一 致で あ る " こ と を 示す。
一方 、 図 4 6 B は 、 時刻 n で の 上述 の マ ク ロ ブ ロ ッ ク の 属性情報 に お け る 隣接画素位置 の ラ ベ ル の 差分 を と っ た結果 を示 し て レ、 る 。 こ こ で 、 左端 の ラ ベルは 、 1 ラ イ ン 上 の右端 の 画素位置 で の ラ ベル と の 差 を 取 り 、 左上端 の 画素位量で の ラ ベルは 、 " 0 " と の 差 を 取 る こ と に し て レ、 る 。 以後 、 便宜的 に 図 4 6 A を フ レ ー ム 間符 号化 、 図 4 6 B を フ レ ー ム 内 符 号化 i n t r a 3 r a m e coding ) と 呼ぶ :: : と にす る 。
図 4 6 A お よ び 4 6 B よ り 、 フ レ ー ム 間符 号化
( inter r r am e coding ) の 方が フ レ - - ム 内 符 号化 に 比べて " S ,, の割合 が 多 く 、 フ レ ー ム 間符 号化 の方 が 予測 が 当 た る た め 、 符 号量 の 削減 を 図 る こ と が で き る な お 、 フ レ ー ム 間 で の 相 関 が極端 に 小 さ い場合 、 フ レ ー ム 内 符 号化 に比べて 符 号化効率が低下す る お そ れ 力 S あ る 。 こ の 場合 は 、 1 ビ ッ ト の符 号で フ レ ー ム 内 符 号化 を行 Ό か 、 フ レ 一 ム 間符 号化 を行 う 力 を 切 り 換 え ら れ る よ う に し て お き 、 フ レ ー ム 内 符 号化 で符 号化 で さ る よ う にす る 。 当 然 の こ と な 力 S ら 、 最初 に符 号化す る フ レ ー ム は 、 参照す る ラ ベ ル が 無 い た め 、 フ レ ー ム 内符 号化 を仃 ぅ 。 こ の 際 、 フ レ ー ム 間 フ レ ー ム 内 を 切 り 換 え る 符 号 は必要 な い
幾種類 力 の 予測 の方法 を 切 り 替 え る 例 を 、 さ ら に 具 体的 に説明す る
上述 の例 で は 、 フ レ ー ム 間 の 相 関 が 小 さ い場合 に 、 フ レ ー ム 内符 号化 を 行 う 場合 を 示 し た。 し カゝ し 、 こ の 他 に も 、 例 え ば本発 明 を 画像伝送 に用 い る 場合 な ど で 伝送誤 り が 問題 と な る 時 に も 、 フ レ ー ム 内符 号化 は有 効 で あ る
例 え ば 、 伝送誤 り が発 生 し 、 前の フ レ ー ム が 正 し く 再生 さ れて い な い場合 に 、 フ レ ー ム 間符 号化 を用 レ、 る と 、 現 フ レ 一 ム も 正 し く 再生 さ れな い が 、 フ レ ー ム 内 符 号化 を 用 い る と 、 正 し く 再生す る こ と 力 S で き る 。 ま た 、 フ レ ー ム 内 符 号化 で あ っ て も 、 多 く の マ ク ロ ブ ロ ッ ク を 参照 し て レヽ る と 、 伝送誤 り に は弱 く な る 。 つ ま り 、 参 照す る マ ク ロ ブ ロ ッ ク を 多 く す る と 、 符 号 量 は少 な く で き る が 、 参照 し た マ ク ロ プ ロ ッ ク の 数が 多 く な れば、 そ れだ け伝送誤 り を含 ん だマ ク ロ プ ロ ッ ク を 参 照す る 可能性 が 高 く な り 、 そ の 参 照 し た マ ク ロ プ ロ ッ ク に含 ま れ る 誤 り が 取 り 込 ま れて 処理結果 に反 映 さ れて し ま う こ と に な る 力ゝ ら 、 伝送誤 り に は弱 く な る と 云 え 、 逆 に 参 照す る マ ク ロ ブ ロ ッ ク を 少 な く す る と 、 符号量 は多 く な る が 、 前記理 由 か ら 伝送誤 り に は 強 く な る と 云 え る 。
そ こ で 、 伝送誤 り に 強 く 、 符号量 も 少 な く す る 工夫 が 必要 な る 。 そ れ に は次 の よ う す る と 良 い。
一例 と し て 、 伝送誤 り に 強 く 、 符 号量 も 少 な く す る た め に は 、 幾つ か の 予測モ ー ド を 用 意 し て お き 、 そ れ ら を 切 り 替 え て用 レ、 る 方法 が 有効 で あ る 。
予測モ ー ド と し て は 、 例 え ば、
( A ) フ レ ー ム 間 符 号化モ ー ド
( B ) フ レ ー ム 内 符号化モ ー ド
( C ) 同 期 回復 単位 内 符 号化モ ー ド
( D ) 予測 な し の モ ー ド
力 あ る 。
こ こ で 、 " 同期 回 復 単位 " と は 、 既 に説 明 し た よ う に 、 オブ ジ ェ ク ト の 矩形領域 を さ ら に 分割 し た も の 、 す な わ ち 、 矩形の符 号化領域 C A を さ ら に所要 の マ ク ロ ブ 口 ッ ク 単位 で分割 し た も の を 意味 し 、 例 え ば 、 各 同 期 回復 単位 の 符 号量が 等 し く な る よ う に分割 し た り あ る い は 、 所定の 数 の マ ク ロ プ ロ ッ ク を ま と め て 同期 回復 単位 と し た り す る
" 同 期 回復 単位 内 符 号ィヒモ 一 ド " で は 、 参 照 プ 口 ッ ク (参照す る マ ク ロ ブ ロ ッ ク の こ と で 、 自 己 の所属 マ ク ロ ブ 口 ッ ク の 隣接マ ク ロ ブ ロ ッ ク ) 力 S 、 た と え フ レ
— ム 内 で あ っ て も 、 " 同 期 回復 単位 " の 外側 に あ る 場 合 は 、 参 照せず に 、 例 え ば予 め 定 め る ラ ベル を 予測値 と す る 。
こ れ に よ り 、 フ レ ー ム 内 で伝送誤 り が発 生 し た と し て も 、 そ れが 、 " 同 期 回復 単位 " の 外で あ れば、 そ の " 同期 回復 単位 " は正 し く 再生す る こ と 力 S で き る 。 ま た 、 予測 な し と は 、 他 の マ ク ロ ブ ロ ッ ク は全 く 参 照せず に 、 該マ ク ロ ブ ロ ッ ク の ラ ベノレ を符 号化す る も の で あ り 、 誤 り に は こ の モ ー ド が 最 も 強 い。
こ れ ら 複数種 の モ ー ド を 用 意 し て 、 誤 り が発 生す る 頻度 に よ っ て いずれか最適 な一つ を 選択切 り 替 え し て 使用 す る 。 そ し て 、 そ の切 り 替 え は 、 " 同 期 回復 単 位 " 毎 に行 っ て も 良 い し 、 フ レ ー ム 毎 に行 つ て も R レヽ し 、 シ ー ケ ン ス 毎 に行 つ て も 民 い。 ど の モ ー ド で符 号 ィ匕 さ れた か と レヽ ぅ 情報 は符 号化装置 か ら 復 号化装置 に 送 ら れ る よ う にす る 。
ま た 、 別 の モ ー ド と し て 、 符 号化 し ょ う と し て い る 対象 の領域の 、 フ レ ー ム 内 に お け る 占 有位置如 何 に よ つ て 、 符 号化テ ー ブル を 切 り 替 え る 方法 も あ る 。
す な わ ち 、 画像 は一般的 な傾 向 と し て 、 例 え ば、 図 1 8 の よ う に フ レ ー ム の 中 心部分 で は 、 ォブ ジ ヱ ク ト が 存在す る 確率が 高 く 、 フ レ ー ム の端で は ォブ ジ ェ ク ト が 存在 し な い確率 が 高 い。
こ の こ と に着 目 し て 、 フ レ ー ム の 端 に接す る マ ク ロ ブ ロ ッ ク で は " transparent" に短い符 号 を 割 り 当 て た テ ー ブル を用 い 、 そ れ以外の マ ク ロ ブ ロ ッ ク で は opaque I二短い符 号 を 割 り 当 て た テ ー ブル を 用 い る よ う にす る と 、 予測 を 用 い な く て も 、 符 号量 を 減 ら す こ と が で き る 。 こ れが 予測 な し モ ー ド で あ る 。
ま た 、 よ り 単純 に 、 複数の符 号化テ ー ブル を 用 意 し そ の符 号化テ一プル を 切 り 替 え て 用 レ、 る 方法 も あ る 。 こ れ の 切 り 替 え 情報 は 、 例 え ば、 " 同期 回復 単位 " 毎 あ る レヽ は 、 フ レ ー ム 毎 、 シ ー ケ ン ス 毎 に符 号化す る 。
図 4 7 A お よ び 4 7 B は 、 上述 し た処理 を 実現す る た め の 本実施例 の シ ス テ ム構成 を 示す ブ ロ ッ ク 図 で あ り 、 こ の ブ ロ ッ ク 図 を 参 照 し て 処理 の 流れ を 説 明 す る こ の 図 4 7 A お よ び 4 7 B の構成 に お い て 、 破線で 囲 ま れた部分 が 前述 し た 処理 を 実現す る 本実施例 に係 わ る 部分 で あ る 。 図 4 7 A は ア ル フ ァ マ ッ プ符 号化装 置 で あ り 、 ォブ ジ ェ ク ト 領域検 出 回路 3 1 0 、 ブ 口 ッ ク 化 回 路 3 1 1 、 ラ ベル付 け 回 路 3 1 2 、 プ ロ ッ ク 符 号化 回路 3 1 3 、 ラ ベル メ モ リ 3 1 4 、 サ イ ズ変 更 回 路 3 1 5 、 ラ ベル符 号化 回 路 3 1 6 、 多重化 回 路 ( M U X ) 3 1 7 と よ り 構成 さ れて い る 。
こ れ ら の う ち 、 オブ ジ ェ ク ト 領域検 出 回 路 3 1 0 は 入力 さ れた ア ル フ ァ マ ッ プ信 号 を 元 に 、 そ の ア ル フ ァ マ ッ プ信号 に お い て オブ ジ ェ ク ト を含 ん で い る 部分 に つ い て の方形領域 を 検 出 し て 、 そ の 方形領域 の サ イ ズ に 関す る 情報 と 共 に 当 該方形領域の ア ル フ ァ マ ッ プ信 号 を 出 力 す る 。 ブ ロ ッ ク ィヒ 回 路 3 1 1 は 、 こ の方形領 域 の ア ル フ ァ マ ッ プ信 号 を マ ク 口 プ ロ ッ ク イヒす る 回 路 で あ り 、 ラ ベル付 け 回 路 3 1 2 は 、 こ の マ ク ロ プ ロ ッ ク イ匕 さ れた ア ル フ ァ マ ッ プ信 号 につ レヽ て そ の ブ ロ ッ ク 毎 に 、 そ の マ ク ロ ブ 口 ッ ク で の ァ ノレ フ ァ マ ッ プ信 号 内 容 の 属性 ( transparent (透 明 の み) 、 Multi (透 明 と 不透明 の 混合) 、 opaque (不透明 の み) ) を 判 定 し 、 各属性 に対応す る ラ ベル ( " 0 " 、 " 1 " 、 " 3 " ) を割 り 当 て る 回 路 で あ る 。
ブ ロ ッ ク 符 号化 回 路 3 1 3 は 、 ラ ベル力 S " 1 " ( Multi) の 属性 の マ ク ロ ブ ロ ッ ク に つ レヽ て 、 そ の マ ク ロ ブ 口 ッ ク 内 の ア ル フ ァ マ ッ プ信 号 を符号ィ匕す る 回 路 で あ り 、 ラ ベル メ モ リ 3 1 4 は 、 ラ ベル付 け 回 路 3 1 2 よ り 供給 さ れ る ラ ベ ル情報 と ラ ベ ル メ モ リ 出 力線 3 0 2 を介 し て オブ ジ ェ ク ト 領域検 出 回 路 3 1 0 力 ら 与 え ら れ る 領域の サ イ ズ情報 を 蓄積す る と 共 に 、 こ の 蓄積 し た ラ ベル情報 と サ イ ズ情報 を 併せ て サ イ ズ変 更 回 路 3 1 5 に供給す る た め の メ モ リ で あ る 。
サイ ズ変更 回 路 3 1 5 は 、 ラ ベル メ モ リ 3 1 4 よ り 供給 さ れ る 、 時刻 n — 1 の フ レ ー ム の ラ ベ ル情報 と サ ィ ズ情報 と 、 オブ ジ ェ ク ト 領域検 出 回 路 3 1 0 か ら 与 え ら れ る 、 時刻 n の フ レ ー ム の サ イ ズ情報 と 力ゝ ら 、 時 刻 n — 1 の ラ ベル情報 を 時刻 n の サイ ズ に相 当 す る 様 に サイ ズ を 変吏す る 回 路 で あ り 、 ラ ベル符 号化 回 路 3 1 6 は 、 こ の サ イ ズ変 更 さ れた ラ ベル情報 を 予測値 と し て 、 ラ ベル付 け 回 路 3 1 2 よ り 供給 さ れ る ラ ベル情 報 を符号化す る 回 路 で あ る 。
ま た 、 多重化 回 路 3 1 7 は 、 ラ ベル符 号化 回 路 3 1 6 の得た符 号化情報 と 、 ブ ロ ッ ク 符 号化 回 路 3 1 3 よ り 供給 さ れ る 符 号化情報 と 、 オブ ジ ェ ク ト 領域検 出 回 路 3 1 0 カゝ ら 与 え ら れ る サ イ ズ情報 と を 多重化 し て 出 力 す る 回路 で あ る 。
こ の よ う な構成 の 符 号化装置 に お い て 、 信 号線 3 0 1 を介 し て 供給 さ れ る ア ル フ ァ マ ッ プ信号 は 、 ォプ ジ ェ ク ト 領域検 出 回 路 3 1 0 に 与 え ら れ、 こ の オ ブ ジ ェ ク ト 領域検 出 回 路 3 1 0 は 、 こ の ア ル フ ァ マ ッ プ信号 か ら ォブ ジ ェ ク ト を含 む方形領域 を 検 出 す る 。 こ の方 形領域の サ イ ズに 関す る 情報 は信 号線 3 0 2 を 介 し て 出 力 さ れ、 領域 内 部 の ア ル フ ァ マ ッ プ信 号 は 、 ブ ロ ッ ク 化 回路 3 1 1 に供給 さ れ る 。
ブ ロ ッ ク ィ匕 回路 3 1 1 は 、 こ の領域 内 部 の ア ル フ ァ マ ッ プ信号 にっ レヽ て マ ク ロ ブ 口 ッ ク 化す る 。 マ ク ロ ブ ロ ッ ク 化 さ れた ア ル フ ァ マ ッ プ信 号は ラ ベル付 け 回路 3 1 2 と ブ ロ ッ ク 符 号ィ匕 回路 3 1 3 に供給 さ れ る 。 19 ラ ベ ノレ 付 け 回 路 3 1 2 0 で は 、 マ ク ロ ブ ロ ッ ク 毎 の 属 性 ( t ransparent 、 Mult i 、 opaque ) を 判 定 し 、 各 属 性 に 対 応 す る ラ ベ ル ( " 0 " 、 " 1 ,, 、 " 3 " ) を 割 り 当 て る 。 こ の 割 り 当 て ら れ た ラ ベ ノレ 情 報 は 、 ブ ロ ッ ク 符 号 化 回 路 3 1 3 、 ラ ベ ル メ モ リ 3 1 4 お よ び ラ ベ ノレ 符 号 ィヒ 回 路 3
1 6 に 供 給 さ れ る 。
ブ ロ ッ ク 符 号ィ匕 回路 3 1 3 で は 、 ラ ベル " 1 " ( Mult i) の と き 、 マ ク ロ ブ ロ ッ ク 内 の ァ ノレ フ ァ マ ッ プ信号が 符 号化 さ れ、 そ の 符 号化情報 は 多重化 回 路 3 1 7 に供給 さ れ る 。 ラ ベル メ モ リ 3 1 4 に は 、 ラ ベル 付 け 回 路 3 1 2 よ り 供給 さ れ る ラ ベル情報 と ラ ベル メ モ リ 出 力線 3 0 2 を介す る 領域 の サ イ ズ情報 が 蓄積 さ れ、 ラ ベル情報 と サイ ズ情報 を 併せて ラ ベル メ モ リ 出 力線 3 0 3 よ り サ イ ズ変 更 回 路 3 1 5 へ と 供給 さ れ る サイ ズ変更 回 路 3 1 5 で は 、 ラ ベル メ モ リ 出 力線 3
0 3 を 介 し て供給 さ れ る 、 時刻 n — 1 の フ レ ー ム の ラ ベル情報 と サイ ズ情報 と 、 信 号線 3 0 2 を 介 し て 供給 さ れ る 時刻 n の サ イ ズ情報 と 力ゝ ら 、 時刻 n — 1 の ラ ベ ル情報 を 時刻 n の サイ ズ に相 当 す る 様 に サ イ ズ を 変 更 し た ラ ベル情報 を ラ ベル符 号化 回 路 3 1 6 に供給す る , ラ ベル符 号化 回 路 3 1 6 で は 、 サ イ ズ変 更 回 路 3 1 5 よ り 供給 さ れ る ラ ベル情報 を 予測値 と し て 、 ラ ベル 付 け 回路 3 1 2 に供給 さ れ る ラ ベル情報 を 符 号化 し 、 そ の符 号化情報 は多重 回 路 3 1 7 に供給 さ れ る 。 多重 ィ匕 回 路 3 1 7 で は 、 ブ ロ ッ ク 符 号化 回 路 3 1 3 ラ ベ ル符 号化 回 路 3 1 3 と ラ ベ ル符 号化 回 路 3 1 6 よ り 供 給 さ れ る 符 号化情報 と 、 ラ ベ ル メ モ リ 出 力線 3 0 2 を 介 し て供給 さ れ る サ イ ズ情報 と を 多重化 し た 後 、 信 号 線 3 0 4 を介 し て 出 力 す る 。
以 上が符 号化装置 の構成 と 作用 で あ る 。 次 に 復 号化 装置 の構成 と 作用 を 説明 す る 。
図 4 7 B に示すア ル フ ァ マ ッ プ符 号化装置 は 、 分離 ィ匕 回 路 ( D M U X ) 3 2 0 、 ラ ベル復 号化 回 路 3 2 1 、 サ イ ズ変 更 回 路 3 2 2 、 ラ ベル メ モ リ 3 2 3 お よ びブ ロ ッ ク 復 号化 回 路 3 2 4 よ り 構成 さ れ る 。
こ れ ら の う ち 、 分離化 回 路 3 2 0 は 、 信号線 3 0 5 を介 し て 供給 さ れ る 符 号化情報 を 分離す る 回 路 で あ り 、 ラ ベル復 号化 回 路 3 2 1 は 、 サ イ ズ変更 回 路 3 2 2 よ り 供給 さ れ る 、 時刻 n — 1 の ラ ベル情報 の サ イ ズ を 変 更 し た情報 を 予測値 と し て 、 時刻 n の ラ ベル情報 を 再 生す る 回 路 で あ る 。
ま た 、 サイ ズ変 更 回 路 3 2 2 は 、 サ イ ズ変 更 回 路 3 1 5 と 同様の働 き をす る 回 路 で あ っ て 、 ラ ベル メ モ リ 3 2 3 よ り 供給 さ れ る 、 時刻 n — 1 の フ レ ー ム で の ラ ベル情報 と サイ ズ情報 と 、 分離化 回 路 3 2 0 か ら 分離 し て 与 え ら れ る 、 時刻 n の フ レ ー ム の サ イ ズ情報 と 力 ら 、 時刻 n — 1 の フ レ ー ム で の ラ ベ ル情報 を 時刻 n の サ イ ズ に相 当 す る 様 に サイ ズ を 変更す る 回 路 で あ り 、 ラ ベル メ モ リ 3 2 3 は 、 ラ ベル メ モ リ 3 1 4 と 同様 の 働 き をす る 回 路 で あ っ て 、 ラ ベル複 号化 回 路 3 2 1 よ り 復 号化 れて 供給 さ れ る ラ ベル情報 と 、 分離化 回 路
3 2 0 力、 ら 与 え ら れ る 領域 の サイ ズ情報 を 蓄積す る と 共 に 、 ^ ~ の 蓄積 し た ラ ベル情報 と サイ ズ情報 を 併せて サ ィ ズ変 更 回路 3 2 2 に供給す る た め の メ モ リ で あ る ま た 、 ブ ロ ッ ク 復 号化 回 路 3 2 4 は 、 ラ ベル復 号化 回 路 3 2 よ り 供給 さ れ る 、 再生 さ れた ラ ベル情報 に 従 つ て 、 ブ 口 ッ ク 毎 に ア ル フ ァ マ ッ プ信 号 を 再生す る 回 路 で あ る
_ の よ な構成 の 復 号化装置 の 作用 を 説 明 す る 。 分離化 回路 3 2 0 で は 、 信 号線 3 0 5 を 介 し て 供給 さ れ る 符 号化情報 を 分離 し て 、 ブ ロ ッ ク 復 号化 回 路 3
2 4 と ラ ぺル複号化 回 路 3 2 1 に供給す る と 共 に 、 信 号線 3 0 6 を介 し て サ イ ズ情報 を 出 力す る 。 ラ ベル復 号化 回 路 3 2 1 で は 、 サ ィ ズ変更 回 路 3 2 2 よ り 供給 さ れ る 時刻 n — 1 の フ レ ー ム で の ラ ベル情報 の サ イ ズ を 変 更 し た情報 を 予測値 と し て 、 時刻 n の フ レ 一 ム で の ラ ベ ノレ 報 を 再生す る 。
再生 さ れた ラ ベル情報 は ブ ロ ッ ク 復 号化 回 路 3 2 4 と ラ ベル メ モ リ 3 2 3 に供給 さ れ る 。 プ ロ ッ ク 復 号化 回 路 3 2 4 で は 、 ラ ベノレ復 号化 回路 3 2 1 よ り 供給 さ れ る 、 再生 さ れた ラ ベル情報 に従 っ て 、 ブ ロ ッ ク 毎 に ァ ル フ ァ マ ッ プ信 号 を 再生す る 。 な お 、 サ イ ズ変 更 回 路 3 2 2 はサ イ ズ変 更 回 路 3 1 5 と 、 ラ ベ /レ メ モ リ 3
2 3 は 、 ベル メ モ y 3 1 4 と 、 各 々 同 一 の 動作 を す る た め 、 こ で は詳 し く 説 明 し な い。 以 上 、 マ ク ロ ブ ロ ッ ク 単位 に し た ア ル フ ァ マ ッ プ を ラ ベル付 け し 、 既 に符 号ィヒ済 みの フ レ ー ム の マ ク ロ ブ ロ ッ ク の ラ ベノレ を利用 し て 、 現 フ レ ー ム の マ ク ロ プ ロ ッ ク の ラ ベル を符号化す る よ う に し た符 号化装置 と 復 号化 装置 の例 を説 明 し た。 時間 的 に 近い フ レ ー ム の ア ル フ ァ マ ッ プ間 で は 、 そ の マ ク ロ ブ ロ ッ ク は非常 に 似 た ラ ベル付 け が 行 われ る 。 従 っ て 、 こ の よ う な 場合 に は 、 フ レ ー ム 間 に お い て ラ ベルの 相 関 が 高 い た め 、 既 に符 号化済 みの フ レ ー ム の ラ ベル を利用 し て 、 現 フ レー ム の ラ ベル を 符 号化す る こ と で 、 大幅 に符 号化効 率 を 図 る こ と が で き る よ う に な る 。
と こ ろ で 、 こ の よ う な 先行技術 と し て の発 明 で は 、 フ レー ム 内 あ る レヽ は フ レ ー ム 間 に お レヽ て 隣接す る 1 ブ ロ ッ ク ( 1 マ ク ロ プ ロ ッ ク ) を 参 照 し て V L C ( 可変 長符 号化) テ ー ブル を 切 り 換 え て い た。 こ の 場合 、 フ レ ー ム 間相 関 が 高 い場合 に は 、 " フ レ ー ム 間 の 隣接 ブ ロ ッ ク " を 参照 し 、 フ レ ー ム 間相 関 が低 レヽ場合 に は 、 " フ レ ー ム 内 の 隣接 ブ ロ ッ ク " を 参 照 し て 、 V L C テ 一ブル を切 り 換 え て い た。 し か し 、 実際 の応用 で は 、 フ レ ー ム 間相 関 と フ レ ー ム 内相 関 両方 を利用 し た方 力 S 良 い場合が 多い。
そ こ で 、 あ る 画素位置 の モ ー ド を " M ( h 、 v 、 t ) " ( h 、 v 、 t は各 々 、 水 平 、 垂直 、 時 間方 向 の 座標軸 を 表す) と す る と 、 モ ー ド " M ( x 、 y 、 n ) " を 符 号化す る 際 に 、 例 え ば、 " M ( X — 1 、 y 、 n ) ,, 、 " M ( x 、 y — 1 、 n ) " 、 " M ( x 、 y 、 n — 1 ) " を 参照 し て V L C テ ー ブル を 選択す る こ と にす る 。 こ こ で 、 図 4 3 A お よ び 4 3 B の よ う に モ 一 ド の数が 3 通 り の場合 、 参 照 プ ロ ッ ク 数が 3 プ ロ ッ ク ( 3 マ ク ロ ブ ロ ッ ク ) な ら ば、 V L C テ ー ブルの 数 は
3 の 3 乗 ( = 2 7 ) 個 と な る 。 ま た 、 参 照 ブ ロ ッ ク 数 を こ れ以 上多 く す る こ と も 可能 で あ る (例 え ば、 " M ( X — 1 、 y — 1 、 n ) 、 M 、 x 、 y 、 n ― 2 ) ,, ) 。
こ の 場合 、 V L C テ ー ブルの 数 が 多 く な る だ け で な く 、 新た に追加す る 参照 ブ ロ ッ ク 間 と の ブ ロ ッ ク 間相 関 も 低下す る た め 、 参 照 ブ ロ ッ ク 数 を 増や し て も 符 号 ィ匕効率 は あ ま り 向 上 し な レ、 。 従 っ て 、 V L C テ ー ブル の数 と 、 符 号化効率 と の ト レ ー ドオ フ を 図 る 必要 が あ る 。
次 に 、 ブ ロ ッ ク の 属性情報 を符 号化す る 別方式の 実 施例 を説 明 す る 。
こ こ で は 、 前の フ レ ー ム の ラ ベル を 予測 に用 い て ブ 口 ッ ク の 属性情報 を符 号化す る 方式 を 説明 す る 。
図 4 8 は本発 明 の一実施例 と し て の 符 号化 回路 の ブ ロ ッ ク 図 で あ る 。 図 に示す よ う に 、 こ の 符 号化 回 路 は オブ ジ ェ ク ト 領域検 出 回路 7 0 2 、 ブ ロ ッ ク ィ匕 回 路 7 0 4 、 ラ ベル付 け 回 路 7 0 6 、 ラ ベル符 号化 回 路 7 0 8 、 ラ ベル メ モ リ 7 0 9 、 参照 ブ ロ ッ ク 決定回 路 7 1 0 、 予測 回 路 7 1 2 と 力 ら 構成 さ れ る 。 こ れ ら の う ち 、 オブ ジ ェ ク ト 領域検 出 回 路 7 0 2 は ア ル フ ァ マ ッ プ の 信 号 7 0 1 力ゝ ら ォブ ジ ェ ク ト を 含 む プ ロ ッ ク サ イ ズの倍数 で表 さ れ る 領域 を 符 号化領域 と し て設 定す る と 共 に 、 符 号化領域 の ア ル フ ァ マ ッ プ信 号 7 0 3 を 切 り 出 す 回 路 で あ り 、 ブ ロ ッ ク 化 回 路 7 0 4 は 、 こ の切 り 出 さ れた ア ル フ ァ マ ッ プ信号 7 0 3 を 1 6 X 1 6 画素構成 の ブ ロ ッ ク 単位 ( マ ク ロ ブ ロ ッ ク 単位) に 分割 ( ブ ロ ッ ク ィ匕) し て 出 力 す る も の で あ り ラ ベル付 け 回 路 7 0 6 は 、 ブ ロ ッ ク 化 さ れた ア ル フ ァ マ ッ プ信 号 7 0 5 に つ い て 、 オ ブ ジ ェ ク ト の含 み具合 に応 じ た所定の ラ ベル を付与 し 、 ラ ベル情報 7 0 7 と し て 出 力 す る も の で あ る 。
ラ ベ ル符号化 回 路 7 0 8 は 、 与 え ら れ る 予測値 7 1 4 に よ っ て符 号化テ ー ブル を 切 り 替 え て ラ ベル情報 7 0 7 を符 号ィヒ し 、 出 力 す る も の で あ り 、 ラ ベル メ モ リ 7 0 9 は 、 ラ ベル付 け 回 路 7 0 6 に よ っ て プ ロ ッ ク 毎 に付与 さ れた前記 の ラ ベル情報 7 0 7 を 記憶す る も の で あ り 、 参 照 ブ ロ ッ ク 決定回 路 7 1 0 は 、 前 フ レ ー ム に お い て 符 号化 プ ロ ッ ク と 同 じ位置 に あ る プ ロ ッ ク を 参 照 ブ ロ ッ ク 7 1 1 と し て 決定す る と い つ た処理 を 行 う も の で あ り 、 予測 回 路 7 1 2 は ラ ベル メ モ リ 7 0 9 に保持 さ れて い る 前 フ レ ー ム の ラ ベル 7 1 3 を 参 照 し . 参照 ブ ロ ッ ク 7 1 1 の位置 の ラ ベル を 予測 し て こ れ を 予測値 7 1 4 と し て ラ ベ ル符 号化 回 路 7 0 8 に 送 る も の で あ る 。 こ の よ う な構成 の符 号化装置 に お い て 、 ア ル フ ァ マ ッ プの 信 号 7 0 1 は 、 ォブ ジ ヱ ク ト 領域検 出 回 路 7 0 2 に入力 さ れ る 。 オブ ジ ェ ク ト 領域検 出 回路 7 0 2 で は 、 オブ ジ ェ ク ト を 含む 、 プ ロ ッ ク サ イ ズ の倍数で表 さ れ る 領域が符 号化領域 と し て 設 定 さ れ、 符 号化領域 で切 り 出 さ れた ア ル フ ァ マ ッ プ 7 0 3 力 S プ ロ ッ ク イ匕 回 路 7 0 4 に 送 ら れ る 。 ブ ロ ッ ク 化 回 路 7 0 4 で は 、 ァ ル フ ァ マ ッ プ 7 0 3 力 S 1 6 X 1 6 画素構成 の ブ ロ ッ ク 単位 ( マ ク ロ ブ ロ ッ ク 単位) に 分割 さ れ、 ブ ロ ッ ク ィ匕 さ れた ァ ノレ フ ァ マ ッ プ 7 0 5 は 、 ラ ベル付 け 回 路 7 0
6 に 送 ら れ る 。 ラ ベル付 け 回 路 7 0 6 で は 、 例 え ば、 . プ ロ ッ ク 内 に オブ ジ ェ ク ト が含 ま れ な い : " ラ ベル 。 "
• ブ ロ ッ ク 内 の 一 部 に オブ ジ ェ ク ト が 含 ま れ る : " ラ ベル 1 "
• ブ ロ ッ ク 内 の 全て が オ ブ ジ ェ ク ト で あ る : " ラ ベ ノレ 3 "
と い っ た ラ ベ ル情報 7 0 7 ( モ ー ド情報) が ブ ロ ッ ク 毎 に付与 さ れ る 。 ラ ベ ル情報 7 0 7 は ラ ベ ル符 号化 回 路 7 0 8 に 送 ら れ、 ま た ラ ベル メ モ リ 7 0 9 に 記憶 さ れ る 。 こ の ラ ベル メ モ リ 7 0 9 に は そ れま で に符 号化 さ れた ラ ベルが 記憶 さ れて い る 。
一方 、 参照 ブ ロ ッ ク 決定回 路 7 1 0 で は 、 例 え ば前 フ レ ー ム に お レヽ て 符 号化 プ ロ ッ ク と 同 じ位置 に あ る ブ ロ ッ ク 力; 参照 ブ ロ ッ ク 7 1 1 と し て 決定 さ れ 、 予測 回 路 7 1 2 に送 ら れ る 。 予測 回 路 7 1 2 に は 、 ラ ベル メ モ リ 7 0 9 力 ら 前 フ レ ー ム の ラ ベル 7 1 3 も 入力 さ れ, そ の う ち の 参照 ブ ロ ッ ク 7 1 1 の位置 の ラ ベルが 予測 値 7 1 4 と し て ラ ベル符 号化 回路 7 0 8 に 送 ら れ る 。 ラ ベル符 号化 回 路 7 0 8 で は 、 予測値 7 1 4 に よ っ て 符 号化 テ ー ブル を 切 り 替 え て ラ ベル情報 7 0 7 を 符 号 化 し 、 符号 7 1 5 力 S 出 力 さ れ る 。
こ こ で 、 符 号化領域 が 常 に フ レ ー ム と 等 し レ、 時 は 、 参照 ブ ロ ッ ク は一つ に 決 ま る 。 し カゝ し 、 符 号化領域 力 S フ レ ー ム よ り も 小 さ く 、 力、つ 、 前 フ レ ー ム と 現 フ レ ー ム で そ の 符 号化額域 の位置 が 異 な る 場合 は 、 フ レ ー ム の 角 を原点 と す る 座標軸 を用 い る か 、 符 号化領域 の 角 を原点 と す る 座標軸 を 用 い る かで参 照 ブ ロ ッ ク は異 な つ て く る 。
こ の座標軸 の 取 り 扱い に つ い て 詳 し く 説明 す る 。
図 4 9 A お よ び 4 9 B は 、 時刻 n _ 1 と 時刻 n に お け る フ レー ム 画像 F n - 1 , F n と そ れぞれの フ レ ー ム F n - 1 , F n に お け る 符 号化領域 C A の 各 マ ク ロ プ ロ ッ ク の モ ー ド情報 M D の例 で あ る 。
先 の 特願平 8 — 2 3 7 0 5 3 号で は 、 一例 と し て 、 現 フ レ ー ム ( 時刻 n ) 内 の符号化領域 の 原点 V c 0 と 、 前 フ レ ー ム (時刻 n — 1 ) 内 の 符 号化領域 の 原点 V p 0 を 一致 さ せて 、 該 ブ ロ ッ ク の モ ー ド情報 を 符 号化す る 際 に参 照 と す る プ ロ ッ ク を 決定す る 実施例 を 提示 し た。 こ れは 、 符 号化領域 の座標軸 に 基づ い て プ ロ ッ ク の 対応付 け を 行 う も の で あ る 。
こ の場合 、 図 5 O A の よ う に 、 前 フ レ ー ム の 符 号化 領域の右端 あ る い は 下端 を " cut" ま た は " no update" す る こ と で 、 現 フ レー ム の 符 号化領域の サ ィ ズ に 一致 さ せて い る 。
図 4 9 A お よ び 4 9 B の 例 で は 、 符 号化領域の左端 お よ び上端が 変化 し て レ、 る 。 こ の よ う な 場合 、 現 フ レ ー ム の モ ー ド情報 に 対応す る ブ ロ ッ ク は 、 図 5 0 A お よ び 5 0 B に示 さ れ る よ う に斜線部 の プ ロ ッ ク ( 2 1 個) が ー致 し て い な い た め 、 こ の値 を用 い て 符 号化す る と 符 号化効率が 低下す る 恐れが あ る 。
図 4 9 A お よ び 4 9 B の よ う な例 で は 、 現 フ レー ム の原点 F c 0 と 前 フ レ ー ム の 原点 F p 0 を 一致 さ せ て フ レ ー ム の座標軸 上 で最 も 近い プ ロ ッ ク 位置 の プ ロ ッ ク を 参照 プ ロ ッ ク と し た方 力 よ レ、。
フ レ ー ム の座標軸 に 基づ い て 参 照 プ ロ ッ ク を 求 め る と ブ ロ ッ ク は図 5 0 B の様 に な る 。 つ ま り 、 図 4 9 A お よ び 4 9 B の例 で は 、 左端 と 上端 で変化 が あ る た め 図 5 0 B の よ う に左端 と 上端 を " cut" ま た は " no update" す る こ と で 、 現 フ レ ー ム の符 号化領域の サ ィ ズ に 一致 さ せ る こ と に な る 。 こ の 場合 、 現 フ レ ー ム の モ ー ド情報 に対応す る ブ ロ ッ ク は 、 図 5 0 B に示 さ れ る よ う に 、 一致 し て い な レ、 ブ ロ ッ ク は 、 斜線部 の ブ ロ ッ ク ( 3 個 ) だ け に な る 。
即 ち 、 状況 に応 じ て 前 フ レ ー ム の ラ ベル を 、 符 号化 領域 の座標軸 に 基づ い て 変 更 す る か 、 フ レ ー ム の 座標 軸 に 基づ レ、 て 変更す る か を 切 り 替 え る こ と に よ り 、 符 号化効率 の 改善が 図 れ る 。 座標軸 の 決定法 は 、 符 号化 装置側 で最適 な方 を 選択 し て 切 り 替 え 情報 を 送っ て も 良 い し 、 符 号化装置 と 複 号化装置 と で共 に既知 の 情報 を 用 レヽ て 決定 し て も よ い。
図 4 5 A お よ び 4 5 B は 、 符 号化領域の座標軸 を 用 レヽ た方 が 良 い例 で あ る 。 こ の 例 は カ メ ラ を右方 向 に振 つ た場合 の よ う に 、 図 4 5 A の フ レ ー ム F n- 1 力 ら 図 4 5 B の フ レ ー ム F n の よ う に フ レ ー ム ;^ 大 き く 移 り 変 わ っ た こ と を示 し て レ、 る 。 こ の 例 の 場合 、 図 力 ら 明 ら か な よ う に 、 フ レ ー ム 内 で の符 号化領域 の位置 が 大 き く ずれて い る た め 、 フ レ ー ム の座標軸 に 基づい て 参照 プ ロ ッ ク を決定す る の は得策 で は な い。
即 ち 、 現 フ レ ー ム (時刻 n 時点 の フ レ ー ム F n ) と 前 フ レ ー ム ( 時刻 n — 1 時点 の フ レ ー ム F n- 1 ) と で フ レ ー ム 内 で の 符 号化領域 C A の位置 が 大 き く ずれ て レ、 る 場合 に は符 号化領域 の座標軸 を用 い た方 が よ く フ レ ー ム 内 で の符 号化領域の位置 が あ ま り ずれて い な い場合 に は フ レ ー ム の座標 軸 を 用 い た方 力 S よ レヽ 。
現 フ レ ー ム F n と 前 フ レ 一 ム F n-1 と で フ レ ー ム 内 で の符 号化領域の位置 が 大 き く ずれで い る か否か は フ レ ー ム 内 に お け る 符 号化領域 の位置 を表す情報 (ベ ク ト ル ) prev_ref scurr_ref と 符 号化領域 の 大 き さ カゝ ら 判 断で き る 。 す な わ ち 、 ア ル フ ァ マ ッ プ を 併用 す る 動 画像符 号化 装置 に お け る 符 号化デー タ 構成 は規格 に よ り 図 5 1 に 示す よ う に な る 。 す な わ ち 、 符 号化デー タ は符 号化領 域の レイ ヤ 、 マ ク ロ ブ ロ ッ ク M B の レイ ヤ 、 2 ィ直开 状 の レイ ヤ を含 み、 符 号化領域 の レイ ヤ は符 号化領域サ ィ ズ情報 、 符号化領域位置情報 、 符 号化領域サ イ ズ変 換率情報 な ど を含 む。 M B の レ イ ヤ は 、 2 値形状情報 , texture M V 情報 、 多値形状情報 、 t e x t u r e 情報 力 ら な り 、 2 値形状情報 はモ ー ド情報 、 動 き べ ク ト ル情 報 、 サ イ ズ変換率情報 、 ス キ ャ ン方 向 情報 、 2 値符 号 化情報 を含む。
こ れ ら の う ち 、 符 号化領域サ イ ズ情報 は符 号化領域 の サ イ ズ (縦横の 大 き さ ) を 表す情報 を 指 し 、 符 号化 領域位置情報 は符 号化領域 の位置 ( V p 0、 V c 0 の位 置) を表す情報 を 指 し 、 符 号化領域サ イ ズ変換率情報 は符 号化領域単位 で の 2 値画像 の サ イ ズ変換率 ( C R ) 情報 を指す。
ま た 、 M B 符 号化情報 は M B 内 の オブ ジ ェ ク ト を 再 生す る た め の 情報 を 指す。 M B の レイ ヤ に お け る 2 値 形状情報 は M B 内 の 各 画素 が オ ブ ジ ェ ク ト 内 部 か否 か を 表す情報 を指 し 、 texture M V 情報 は M B 内 の輝 度信 号や色差信 号 を 動 き 捕償予測す る た め の 動 き べ ク ト ル情報 を指 し 、 多値形状情報 は ォブ ジ ク ト を 他 の ォブ ジ ェ ク ト と 合成す る 際 の 重みづ け (weighting ) 情報 を 指 し 、 t e X t u r e 情報 は M B 内 の輝度信 号や色 30 差信号の 符 号化情報 を 指す。
ま た 、 2 値形状 の レ イ ヤ に お け る モ 一 ド情報 は M B 内 の 2 値画像 の 属性 を 表す情報 を 指 し 、 動 き べ ク ト ル 情報 は M B 内 の 2 値の 画像 を 動 き 補償予測す る た め の 動 き べ ク ト ル情報 を指 し 、 サ イ ズ変換率情報 は M B 単 位 で の 2 値画像の サ イ ズ変換率 ( C R ) 情報 を 指 し 、 ス キ ヤ ン方 向 情報 は符 号化 の順序 が 水 平方 向 か垂直方 向 か を 示す情報 を 指 し 、 2 値符 号化情報 は 2 値画像 の 符 号化情報 を指す。
符 号化領域の位置 ( V p 0、 V C 0 の位置) を 表す情 報 は 、 符 号化領域位置情報 に納 め ら れて お り 、 従 っ て こ の 情報 を 用 い る こ と で符 号化領域の位置 ( V p 0、 V c O の位置) を 知 る こ と が で き る 。 こ の 情報 を用 い て フ レ ー ム F n- 1 と フ レ ー ム F n を 比較す る 。 こ の 比較 は フ レー ム の ホ ー ム ポ ジ シ ョ ン か ら 符 号化領域 の ホ ー ム ポ ジ シ ョ ン位置 ま で の べ ク ト ル を そ れぞれ求 め て こ の べ ク ト ノレ に よ っ て 行 う 。
そ の 結果 、 例 え ば、 図 4 5 A お よ び 4 5 B の様 に " rev ref と curr r e f " の 差力 大 き く 、 現 フ レ ー ム と 前 フ レ ー ム と で符 号化領域 の 大 き さ が ほ と ん ど変 化 が な い場合 に は 、 符 号化領域 の座標軸 に 基づ い て た 方 良 レヽ こ と が 分 力、 る 。 こ こ で 、 " prev— ref と curr_ref " お よ び符 号化領域 の 大 き さ の 情報 は 、 符 号化領域 の符号化 に先 立 っ て 符 号化 さ れて お り 、 復 号 装置側 で も 既知 の 情報 で あ る た め 、 ど ち ら の座標 軸 を 用 い た か を識別す る 付加情報 は必要 な い。
ま た 、 図 5 2 A に示す よ う に フ レ ー ム 7 3 0 の 一部 分 と し て 符 号化領域 7 3 1 が 設定 さ れた場合 、 符 号化 領域 7 3 1 の 外側 は ラ ベルが 決 ま ら な レ、 。 し 力、 し 、 次 の フ レ ー ム の符 号化 の 際 に は 、 符 号化領域 7 3 1 の外 側 が 参 照 ブ ロ ッ ク と な る 事 が あ る の で 、 何 ら か の ラ ベ ル を 挿入 し て お く 必要 が あ る 。
図 5 2 D は 、 図 5 2 A に お け る ラ ベル未定の 部 分 に 所定の値、 こ の 場合 は " 0 " を 挿入 し た例 で あ る 。 図 5 2 C は 、 図 5 2 A に お け る ラ ベル未 定部分 につ い て 、 符 号化領域か ら 外揷 し た例 で あ り 、 こ れは 、 ォプ ジ ェ タ ト の 動 き が 大 き 力、 つ た り 、 形状 が激 し く 変化す る な ど し て 、 前 フ レ ー ム で はォブ ジ ェ ク ト 力 無 力ゝ つ た 部分 に次 の フ レ ー ム で は オブ ジ ェ ク ト が 現れ る 確率 が 高 い 場合 に 有効 で あ る 。
図 5 2 B は 、 図 5 2 A の ラ ベル未定部分 に つ い て 、 ラ ベル メ モ リ 7 0 9 の メ モ リ 空間 の う ち 、 次 の フ レー ム の 符 号化領域 7 3 2 の 部分 だ け を 外挿 し 、 そ の他の 部 分 は上書 き し な レ、例 で あ り 、 こ の よ う に す る と 、 2 フ レ ー ム 以 上前 の フ レー ム の ラ ベル を 予測 に 用 レ、 る こ と 力 S で き る 。
以 前 の フ レ ー ム の ラ ベル を 予測 に用 い た い場合 は 、 他 に 、 例 え ば外挿や所定値の 挿入 は全 く 行 わず に 、 メ モ リ 空間 の う ち 、 符 号化領域 だ け を 更新 し て い く 方法 も あ る 。 次 に 、 図 3 7 A な レヽ し 3 7 D を 用 レ、 て 説 明 し た 、 フ レ ー ム の サ イ ズ変換率 ( C R ) が フ レ ー ム 毎 に切 り 替 わ る 場合 の ラ ベ ル の 予測方法 に つ い て 説 明 す る 。
図 5 3 は 、 縮小処理 の例 で あ っ て 、 前 フ レ ー ム 力 S " C R = 1 " で現 フ レ ー ム カ S " C R = 1 / 2 " の 例 で あ り 、 こ の 場合 、 縮小 し て 得 よ う と す る 現 フ レ ー ム の 対象 ブ ロ ッ ク で あ る 例 え ばマ ク ロ ブ ロ ッ ク M B 1 に 対 応す る 前 フ レ一ム の ブ 口 ッ ク は 、 図 に 示 し た よ う に 、 M B 2 〜 M B 5 の 4 つ あ る ( 図 5 4 参 照 ) 。 つ ま り 、 前 フ レ — ム の マ ク π ブ π ッ ク M B 2 , M B 3 , M B 4 M B 5 が縮小後 の マ ク 口 ブ ロ ッ ク M B 1 に な る 。
今 、 現 フ レ — ム の マ ク ロ ブ ロ ッ ク M B 1 の ァ ド レ ス を ( X , y ) と す る と 、 前 フ レ ー ム の ブ ロ ッ ク M B 2
〜 M B 5 の ァ ド レ ス は ( 2 X , 2 y に ( 2 + 1 , 2 y ) 、 ( 2 X , 2 y + 1 ) 、 ( 2 x + l 、 2 y + 1 ) で得 ら れ る 。
で の係数 " 2 " は 、 前 フ レ ー ム と 現 フ レ ー ム の サイ ズ変換率 C R の値の 比 と し て 与 え ら れ る 。
プ 口 ッ ク M B 1 の ラ ベル の 符 号化 の 予測 に は 、 ブ ロ ッ ク M B 2 、 M B 3 、 M B 4 、 M B 5 の いずれ力 の ラ ベ ノレ 用 い る の が 適 当 で あ る が 、 そ の 決 め 方 に は い く つ か方法が あ る 。
ま ず 、 最 も 簡 単 で演算 量 が 少 な レヽ の は 、 4 つ の う ち の 所定の位置 (例 え ば左 上) に あ る ブ ロ ッ ク の ラ ベノレ を 用 い る 方法 で あ る 。 あ る レ、 は 、 4 つ の ラ ベル に 同 じ も の が あ る 時 は 、 そ の 数が 最 も 多 い ラ ベ ル を 予測値 と し て 用 い る よ う にすれば、 予測 力 S 当 た る 確率 は 高 く な る 。
同 じ 内 容 を 持つ ラ ベ ル の 数が 等 し い場合 、 つ ま り 、 二組ずつ に分 かれて し ま っ た場合 に は 、 出 現頻度 が 高 い順 に 予 め ラ ベル に順序 を 付 け て お き 、 そ の順序 で 上 位 の ラ ベ ル を 予測値 と し て 選択す る 。 フ レ ー ム 内 に矩 形 の 符 号化領域 C A を 設 定す る 場合 、 符 号化領域 C A の 角 を 原 点 と す る 座標軸 を 用 レ、 る と 、 図 5 4 の よ う に 現 フ レ ー ム の ブ ロ ッ ク (符 号化対象 の マ ク ロ ブ ロ ッ ク ) の 境界 は参 照 フ レ ー ム の マ ク ロ ブ 口 ッ ク の 境界 に 重 な る
し か し 、 符 号化対象 の領域 はマ ク ロ ブ 口 ッ ク の 幅 よ り も 小 さ な ス テ ッ プ で位置設 定可能 と し 、 フ レ ー ム の 角 を原 点 と す る 座標軸 を 用 い る と 、 図 5 5 の よ う に 、 一般 に ブ ロ ッ ク の 境界 は重 な ら ず 、 M B 6 〜 M B 1 4 の 計 9 つ の マ ク ロ ブ 口 ッ ク カ S 参 照 さ れ る 。
こ の場合は 、 全体 が 参照 さ れ て い る マ ク ロ ブ 口 ッ ク M B 1 0 の ラ ベル を 用 い る 。
図 5 6 は 、 前 フ レ ー ム が サ イ ズ変換率 C R = 1 / 2 で 現 フ レ ー ム が サ イ ズ変換率 C R = 1 の 例 で あ り 、 こ の 場合 、 マ ク ロ ブ ロ ッ ク M B 1 9 は縮小 フ レ ー ム 中 の マ ク ロ ブ ロ ッ ク M B 1 5 の右 下 の 部分 を 参 照す る こ と に な る 。 こ の 時 は 、 プ ロ ッ ク M B 1 5 の ラ ベル を 予測 値 と す る 力 あ る レ、 は 、 参 照 部分 ブ ロ ッ ク M B 1 6 〜 M B 1 8 と も 近 レヽ こ と 力 ら 、 こ れ ら の プ ロ ッ ク の ラ ベル を 考慮 に入れて 、 例 え ば前述 し た よ う に 多数決 な ど で予測値 を 決 め て も よ い。
図 5 7 に ラ ベル を 予測 に用 レヽ る よ う に し た本発 明 の 復 号化装置 の構成例 を ブ ロ ッ ク 図 で示す。
こ の複 号化装置 は 、 ラ ベル複 号化 回 路 7 1 6 、 ラ ベ ル メ モ リ 7 1 7 、 参 照 ブ ロ ッ ク 決定回 路 7 1 8 、 予測 回 路 7 2 0 と カゝ ら 構成 さ れ る 。
こ れ ら の う ち 、 ラ ベ ル複 号化 回 路 7 1 6 は 、 入力 さ れた復 号対象 の符 号デー タ か ら ラ ベルの デー タ を 復 号 ィ匕す る も の で あ り 、 ラ ベ ル メ モ リ 5 1 7 は 、 こ の 復 号 ィ匕 さ れた ラ ベルの デー タ を 記憶す る も の で あ り 、 参 照 ブ ロ ッ ク 決定回 路 7 1 8 は 、 前 フ レ ー ム に お い て 符 号 ィ匕 ブ 口 ッ ク と 同 じ位置 に あ る プ ロ ッ ク を 参 照 プ ロ ッ ク 7 1 9 と し て 決定す る と い つ た 処理 を 行 う も の で あ る ( ま た 、 予測 回 路 7 2 0 は前 フ レ ー ム の ラ ベ ル 7 2 1 と 参 照 ブ ロ ッ ク 7 1 9 力 ら 予測値 7 2 2 を 求 め 、 ラ ベ ル復 号化 回 路 7 1 6 に 与 え る 機能 を 有す る 。
こ の よ う な構成 の 複 号化装置 に お い て 、 復 号対象 で あ る 符 号化デー タ の ス ト リ ー ム 7 1 5 は ラ ベ ル複 号化 回 路 7 1 6 に入力 さ れ、 ラ ベルが 復 号化 さ れ る 。
一方 、 ラ ベル メ モ リ 7 1 7 に は 、 そ れま で に複 号化 さ れた ラ ベルが 記憶 さ れて い る 。 ま た 、 参 照 ブ ロ ッ ク 決定回 路 7 1 8 で は 、 符 号化装置 で説 明 し た も の と 同 じ ょ う に参 照 プ ロ ッ ク 7 1 9 が 決定 さ れ予測 回 路 7 2 0 に 送 ら れ る 。 予測 回 路 7 2 0 で も 符 号化装置 と 同様 に 前 フ レー ム の ラ ベノレ 7 2 1 と 参 照 ブ ロ ッ ク 7 1 9 力 ら 予測値 7 2 2 が 求 め ら れ、 ラ ベル復 号化 回 路 7 1 6 に送 ら れ る 。 ラ ベル復 号化 回路 7 1 6 で は 、 予測値 7
2 2 に よ つ て 復 号化 テ ― ブノレ を 切 り 替 え て ラ ベル 7 2 3 を 復 号化 し 出 力 す る
ま た 、 あ る ブ ロ ッ ク の モ ー ド " M ( h 、 V 、 t ) ,, ( h 、 V 、 t は各 々 、 水 平 、 垂直 、 時 間 方 向 の座標軸 を 表す) を符 号化す る 際 に 、 例 え ば、 " M ( X — 1 、 y 、 n ) M ( 、 y — 1 、 n ) " 、 " M ( X 、 y 、 n - 1 ) ,' な ど を 参 照 し て 符 号化 テ ー ブ ル を 切 り 替 え る の で あ る が、 こ こ で用 い る モ ー ド に 、 動 き 補償 に用 い る 動 き べ ク ト ルの 情報 の一部 を含 め 、 以 下 に示 すモ ー ド の集合 ( こ れ を モ ー ド集合 A と 呼ぶ こ と にす る 。 ) を用 い る こ と ち で き る 。
[ モ一 ド集合 A ]
( 1 ransparen
( 2 ) opaque
( 3 ) no update (動 き ベ ク 卜 ル
0 )
( 4 ) " no update (動 き べ ク ト ル ! =
0 ) ,,
( 5 ) coded
こ こ で 、 モ ー ド集合 A の ( 3 ) と ( 4 ) は いずれ も co y モ一 - ド で あ る が 、 モ ー ド集合 A の ( 3 ) は 、 動 き べ ク ト ルが ゼ ロ で あ る こ と を 意味 し 、 モ ー ド集合 A の ( 4 ) は 、 動 き べ ク ト ルが ゼ ロ 以外で あ る こ と を 意 味す る 。 モ ー ド集合 A の ( 4 ) の 場合 は動 き べ ク 卜 ル の値 を別途符 号化す る 必要 が あ る が 、 モ一 ド集合 A の ( 3 ) の 場合 は 、 動 き べ ク ト ル を 符 号化す る 必要 は な レヽ 。 動 き べ ク ト ル ; ^ ゼ 口 で あ る 確率 が 高 い場合 に 、 モ ― ド集合 A を 用 レ、 る と 、 モ ー ド の 符 号量 と 動 き べ ク ト ルの 符 号量の 合計 を 削 減で き る 。
こ の 例 に ぉ レヽ て 、 " no u date 動 さ べ ク ト ル = = 0 ) " と し て 得 ら れ る プ ロ ッ ク の 全て の 画素 が 例 え ば不透 明 の 場合 、 前記 ( 2 ) と ( 3 ) は ど ち ら の モ ー ド で も 同 じ再生画像が 得 ら れ る 。 つ ま り 、 こ れ ら 2 の モ ー ド を 分 け る 必要 は な レヽ。 同様 に 、 co y 動 き べ ク ト ル = = 0 ) " と し て 得 ら れ る ブ 口 ッ ク の 全 て の 画素 が透 明 の 場合 、 前記 ( 1 ) と ( 3 ) は分 け る 必 要 は な い。 そ こ で 、
[ モ ー ド集合 B 」
( 1 ) transparent
( 2 ) o aque
( 3 ) " no update 、動 き べ ク ト ル !
0 )
( 4 ) " coded"
と し て
ス テ ッ プ A 1 : " no update 、動 き べ ク ト ル = = 0 ) " " と し た 時 に得 ら れ る 動 き 補償予測画像 が 全て 不透明 で あ る 場合 は 、 ス テ ッ プ A 3 へ 、 そ う で な い場 合 は 、 ス テ ッ プ A 2 へ進 む。
ス テ ッ プ A 2 : " no update (動 き べ ク ト ル = 0 ) " と し た 時 に得 ら れ る 動 き 補償予測画像 が 全 て 明 で あ る 場合 は 、 ス テ ッ プ A 4 へ 、 そ う で な い場合 は 、 ス テ ッ プ A 5 へ進 む。
ス テ ッ プ A 3 : 参 照す る " " M ( * 、 氺 、 * ) " が 、 " no update (動 き べ ク ト ル = = 0 ) " の 場合 ¾ " M ( * 、 * 、 * ) " を " o aque" に . 置 き 換 え る 。 ス テ ッ プ A 6 に進 む。
ス テ ッ プ A 4 : 参 照す る " M ( * 、 * 、 * ) ,, が no update (動 き べ ク ト ル = = 0 ) " の 場合 に は M ( ネ 、 * 、 * ) " を " transparent" に 置 き 換 ん る 。 ス テ ッ プ A 6 に進む。
ス テ ッ プ A 5 : " M ( x 、 y 、 n ) ,, を モ ー ド集
A の 符 号化 テ ー ブル を 用 い て 符 号化す る 符 号化終 了
ス テ ッ プ A 6 : " M ( X 、 y 、 n ) " を モ ー ド集 A の 符 号化 テ ー ブ ル を用 い て 符 号化す る 。 符 号化終 了
以 上 の A 0 力 ら A 6 の 手順 を 踏む ア ル ゴ リ ズ ム ( 図
5 8 ) を 用 い れば、 同 じ結果 と な る も の に複数の モ 一 ド、が 用 意 さ れ る と い う 無駄 が 無 く な り 、 ブ ロ ッ ク の 属 性情報の符号量 を 削減で き る 。 な ぜ な ら 、 4 つ の モ ー K (モ ー ド集合 B ) を切 り 替 え る 符 号の 平均符 号長 は 38
5 つ の モ ー ド ( モ ー ド集合 A ) を 切 り 替 え る 符 号の 平 均符 号長 よ り も 短 く で き る カゝ ら で あ る 。 但 し 、 モ ー ド 集合 A の み を ブ ロ ッ ク 毎 に 切 り 替 え る 方式 は 、 モ ー ド 集合 A の み を 用 レ、 る 場合 と 比較 し て 、 演算 量 と メ モ リ 量が 多少増加す る の で 、 こ の増加 が 問題 に な ら な い場 合 に用 レ、 る 。
複 号化処理 は 図 5 8 の フ ロ ーチ ャ ー ト と 全 く 同 じ ァ ル ゴ リ ズ ム で符 号化 テ ー ブ ル を モ ー ド集合 A 用 あ る い はモ ー ド集合 B 用 の いずれか に決定 し 、 そ の テ ー ブル を 用 い て 復号化す る 。
上述 の ア ル ゴ リ ズ ム と 同 じ効果が得 ら れ る 別 の ア ル ゴ リ ズム を 図 5 9 に示す。 こ こ で 、
[ モ ー ド集合 C ]
( 1 ) transparent
( 2 ) " no update (動 き べ ク ト ル = =
0 ) "
( 3 ) no u D d a t e (動 ク ト ル
0 )
4 ) coded
[ モ一 ド集合 D ]
丄 ) opaque
2 " no u date (動 さ ク ト ル =
0 )
no update 動 き べ ク ト ル ! =
0 ) ( 4 ) " coded"
で あ る 。
図 5 9 の フ ロ ー チ ャ ー ト を説 明 す る 。
ス テ ッ プ B 1 : " no update (動 き べ ク ト ノレ = = 0 ) " と し た 時 に得 ら れ る 動 き 補償予測 画像 が 全て 不 透明 で あ る 場合 は 、 ス テ ッ プ B 3 へ 、 そ う で な い場合 は 、 ス テ ッ プ B 2 へ進む。
ス テ ッ プ B 2 : " no update (動 き べ ク ト ノレ = = 0 ) " と し た 時 に得 ら れ る 動 き 補償予測 画像 が 全て 透 明 で あ る 場合 は 、 ス テ ッ プ B 4 へ 、 そ う で な い場合 は 、 ス テ ッ プ B 5 へ進 む。
ス テ ッ プ B 3 : 参 照す る " M ( * 、 * 、 * ) " " カ 、 " opaque" の 場合 に は " M ( * 、 * 、 * ) " を " no update (動 き べ ク ト ル = = 0 ) " に 置 き 換 え る 。 ス テ ツ プ B 6 に進む。
ス テ ッ プ B 4 : 参 照す る " M ( * 、 * 、 * ) " 力 S 、 " transparent" の 場合 に は " M ( * 、 * 、
* ) " を " " no update (動 き べ ク ト ノレ = = 0 ) " に 置 き 換 え る 。 ス テ ッ プ b 7 に進 む。
ス テ ッ プ B 5 : " M ( X 、 y 、 n ) ,, を モ ー ド集合
A の符 号化テ ー ブル を 用 い て符 号化す る 。 符 号化終 了 。
ス テ ッ プ B 6 : " M ( x 、 y 、 n ) " を モ ー ド集合 C の符号化テ ー ブル を用 い て 符 号化す る 。 符 号化終 了 。
ス テ ッ プ B 7 : " M ( x 、 y 、 n ) ,, を モ ー ド集合 D の 符 号化 テ ー ブル を用 い て 符 号化す る 。 符 号化終 了 。 ま た 、 ブ ロ ッ ク の 属性 と し て は 、 符 号化 の パ ラ メ 一 タ 、 例 え ばプ ロ ッ ク サ イ ズや ブ 口 ッ ク の 縮小率 、 符 号 化 ス キ ャ ン の方 向 、 動 き べ ク ト ル の値 な ど も 必要 に応 じ て含 め る こ と も で き る 。 ス キ ャ ン方 向 を含 め た例 を 示す
Cモ ー ド集合 E ]
、 1 ) transparent
2 ) o aque
( 3 ) " no update (動 さ べ ク ト ル
0 )
( 4 ) no u date ( ク 卜 ノレ
0 )
( 5 ) " coded 力 つ水 平 ス キ ャ ン "
( 6 ) " coded 力 つ垂直 ス キ ャ ン ,,
な お 、 " = = " は左辺 は右辺 の値 と 等 し い 、 " ! は左辺 は右辺 の値 と 等 し く な い こ と を 示す。
以 上 、 述べて き た よ う に 、 本発 明 で は 、 前 の フ レ ー ム の 参照 ブ ロ ッ ク を 予測 に 用 い る の で あ る が 、 こ の 参 照 プ ロ ッ ク の ラ ベ ル で は な く 、 ア ル フ ァ マ ッ プそ の も の を 予測 に用 い て も 構わ な い。 す な わ ち 、 ア ル フ ァ マ ッ プ を メ モ リ に記億 し て お き 、 各 ブ 口 ッ ク の 属性 を 符 号化す る 度 に 、 そ の 参照 ブ ロ ッ ク の 属性
( " transparent 、 M u 11 i 、 o aque な ど) を 判 定 し 、 そ の 属性 に よ っ て 符 号化テ ー ブル を 切 り 替 え る 。 こ の よ う にす る と 、 前 の フ レ ー ム を 符 号化 し た 時 に 用 い た プ ロ ッ ク と は 、 位 置 が 数画素ずれた も の も 参 照 ブ ロ ッ ク と す る こ と 力 S で き る 。 つ ま り 、 参 照 ブ ロ ッ ク が 前の フ レー ム を 符 号化 し た 時 に用 い た ブ ロ ッ ク と ピ ッ タ リ と 重 な ら な く て も 力 ま わ な く な り 、 よ り 精度 の 高 い 予測 が 可能 と な る 。
ま た 、 参照 プ ロ ッ ク の ア ル フ ァ マ ッ プ を用 い た 予測 と 、 ラ ベル を 用 い た 予測 を組み合 わせ る こ と も 可能 で め O 。
例 え ば、 ま ず、 参 照 ブ ロ ッ ク の ア ル フ ァ マ ッ プ を 用 レヽ て 、 transparent 、 Multi 、 o aque に よ っ て 符 号化テ ー ブル を 切 り 替 え 、 さ ら に 、
" Multi" の も の に っ レ、 て は 、 参照 ブ ロ ッ ク の ラ ベノレ を 用 い て符号化テ ー ブル を 切 り 替 え る よ う にす る 。
前の フ レ ー ム の 参 照す る 部分 に つ い て は 、 プ ロ ッ ク 毎 に 与 え ら れ る 動 き べ ク ト ル を利用 す る 方法 も あ る 。 つ ま り 、 既 に符 号化 の 終わ っ た 、 符 号化 ブ ロ ッ ク に 隣 接す る マ ク ロ ブ ロ ッ ク の 動 き べ ク ト ル に よ っ て 指 し示 さ れ る 部分 を 前 の フ レ ー ム 力、 ら 切 り 出 し 、 そ れが 、 transparent 力 、 Multi 力 、 o aque" 力 に よ っ て 符 号化テ ー ブル を 切 り 替 え る 。
本発 明 は画像 を 符 号化す る 場合 に 、 画 面 内 を 背景 と オブ ジ ェ ク ト と に 分割 し て 符 号化す る 方式 に お い て 、 背景 と オブ ジ ェ ク ト を 分 け る た め に 、 オ ブ ジ ェ ク ト 形 状や画面 内 の位置 を 表す ァ ノレ フ ァ マ ッ プ信 号 を使用 す る 。 そ し て 、 こ の ア ル フ ァ マ ッ プ信 号 を 画像 の 符 号化 情報 と 共 に符 号化 し て ビ ッ ト ス ト リ 一 ム 化 し 、 伝送す る よ う に し た り 蓄積 し た り す る が 、 前者 は放送ゃパ ソ コ ン通信 に 、 そ し て 後者 は ミ ュ ー ジ ッ ク C D な ど の よ う に 、 コ ン テ ン ツ を 納 め た 商 品 と し て 取 り 引 き さ れ る こ と に な る 。
商 品 と し て 記憶媒体 に記録 し た動画像 コ ン テ ン ツ を 提供す る 場合 に 、 蓄積媒体 に 画像 の 符 号化情報 と こ の ア ル フ ァ マ ッ プ信 号の圧縮符 号化 し た も の と を 合 わ せ た ビ ッ ト ス ト リ ー ム で 蓄積 し て 、 一つ の 媒体 で長 時 間 に 亘 る コ ン テ ン ツ を 納 め て 映画 な ど を鑑賞 で き る よ う にす る が 、 こ の よ う な ア ル フ ァ マ ッ プの 画像 を含 む圧 縮符 号化 し た ビ ッ ト ス ト リ ー ム を 格納 し た 蓄積媒体 を 対象 と す る 再生 シ ス テ ム の 例 を 次 に 第 7 の 実施例 と し て説明 す る 。
図 6 0 お よ び図 6 1 を 用 い て 本実施例 を 説 明 す る 。 図 6 0 は 、 図 1 9 の V L C . 多重化 回路 1 8 0 に お レヽ て 多重化 さ れた 、 モ ー ド情報 ( shape mode ) b 0 N 動 き べ ク ト ル情報 b 1 、 サ イ ズ変換率情報
( convers ion ratio ) b 2 、 ス キ ャ ン方 向 情報
( scan type ) b 3 、 2 値画像符 号化情報 b 4 の 符 号 歹 (J ( bitstreara )の フ ォ 一 マ ツ ト ij を 示 し て レ、 る 。 本 発 明 で は 、 2 値画像符 号化情報 b 4 を復 号す る 際 に 、 b 0 〜 b 3 ま で の 情報 が 復 号 さ れて レ、 る 必要 が あ る 。 ま た 、 モ ー ド情報 b 0 は他 の全て の 情報 b 1 〜 b 4 に 先 だ っ て 復号 さ れて い な い と 、 他 の 情報 を 復 号す る こ と は 出 来 な い。 従 っ て 、 各情報 b 0 〜 b 4 は 、 図 6 0 の よ う に 、 先頭 にモ ー ド情報 、 最後 に 2 値面像符 号化 情報 が 配置 さ れ る 構成 で な け れ ばな ら な い。
図 6 1 は 、 図 6 0 で示 さ れ る 符 号列 が 蓄積 さ れ る 記 録媒体 8 1 0 を用 い て 画像信 号 を 再生す る シ ス テ ム を 示す図 で あ る 。 記録媒体 8 1 0 に は 、 図 6 0 で示 さ れ た符 号列 を含 む符 号列 が 蓄積 さ れて い る 。 デ コ ー ダ 8 2 0 は こ の 蓄積媒体 8 1 0 に 蓄積 さ れて い る 符 号列 か ら 、 画像信号 を 再生す る 。 画像情報 出 力 装置 は 8 3 0 は再生画像 を 出 力 す る 。
こ の よ う な構成 の 本 シ ス テ ム は 、 蓄積媒体 8 1 0 に 図 6 0 に示す如 き の フ ォ ー マ ツ ト の符 号列 を 蓄積 し て あ る 。 デ コ ー ダ 8 3 0 は こ の 蓄積媒体 8 1 0 に 蓄積 さ れて い る 符 号列 か ら 、 画像信 号 を 再生す る 。 す な わ ち デ コ ー ダ 8 2 0 は 、 蓄積媒体 8 1 0 よ り 信 号線 8 0 1 を介 し て 符 号列 を 読み込 み 、 図 6 2 お よ ぴ図 6 3 に 示 す手順 に よ り 再生画像 を 生成す る 。 こ こ で 、 図 6 3 は 図 6 2 の " 2 ィ直画像再生 " の ス テ ッ プ ( S 5 ) の フ ロ — チ ャ ー ト で あ る 。
図 6 2 お よ び図 6 3 に従 っ て 、 デ コ ー ダ 8 2 0 で の 処理 を 内 容 を 説 明 す る 。 す な わ ち 、 ま ず は じ め に モ ー ド情報 が 復 号 さ れ ( ス テ ッ プ S 1 ) 、 そ の 復 号 さ れた モ ー !" 情報力 transparent , o p a a u e " , " no update" の いずれか に該 当 す る か否 か を 調べ る ( ス テ ツ プ S 2 , S 3 , S 4 ) 。
そ の結果 、 " transparent" で あ れ ばマ ク ロ プ ロ ッ ク 内 の 画素値 を 全て 透 明 に し て 処理 を 終 え ( ス テ ツ プ S 6 ) 、 ま た 、 " transparent" で な く 、
" o aq e" で あればマ ク ロ ブ ロ ッ ク 内 の 画素ィ直 を 全 て 不透明 に し て 処理 を 終 え ( ス テ ッ プ S 7 ) 、 ま た 、 transparent で t> な \ 、 o aque で も な く 、 " no u date" で あ っ た と すれ ば動 き べ ク ト ノレ を 'If 報 を 復 号 し ( ス テ ッ プ S 8 ) 、 動 き 補償予測 し ( ス テ ッ プ S 9 ) 、 マ ク ロ ブ ロ ッ ク 内 を こ の 求 め た動 き 補償 予測値で コ ピー し ( ス テ ッ プ S 1 0 ) 、 処理 を 終了 す る 。
一方 、 ス テ ッ プ S 2 , S 3 , S 4 に お い て
transparent で も な \ 、 o aque で t> な く 、 " no update" で も な カゝ つ た場合 に は 2 値画像再生 の 処理 に移 る ( ス テ ッ プ S 5 ) 。
こ の S 5 で の 処理 は 図 6 3 に示す如 き で あ り 、 ま ず は じ め に " inter" 符 号化 で あ る か否 か を 調べ る ( ス テ ツ プ S 2 1 ) 。 そ の 結果 、 " inter" 符号化 で あ れ ば動 き べ ク ト ル情報 を 復 号 し ( ス テ ッ プ S 2 5 ) 、 動 き 補償予測 し ( ス テ ッ プ S 2 6 ) 、 次 に サイ ズ変換率 情報 を 復 号 し ( ス テ ッ プ S 2 2 ) 、 ス キ ャ ン方 向 情報 を 復 号す る ( ス テ ッ プ S 2 3 ) 。 2 値符 号化情報 を 復 号 し ( ス テ ッ プ S 2 4 ) 、 処理 を 終 了 す る 。
一方 、 ス テ ッ プ S 2 1 で の 判 断の 結果 、 " inter" 符 号化 で な か っ た と き は サ イ ズ変換率情報 を 復 号 し ( ス テ ッ プ S 2 2 ) 、 ス キ ャ ン方 向 情報 を 復 号す る ( ス テ ッ プ S 2 3 ) 。 2 値符 号化情報 を 復 号 し ( ス テ ッ プ S 2 4 ) 、 処理 を 終了 す る 。
こ の よ う 〖こ し て デ コ ー ダ 8 2 0 で は画像 を 再生 し 、 こ の 再生 し た 面像 を 信 号線 8 0 2 を介 し て 画像情報 出 力 装置 8 3 0 に供給す る こ と に よ り 、 当 該再生 面像 が こ の 画像情報 出 力 装置 8 3 0 に表示 さ れ る 。
こ こ で 、 画像情報 出 力 装置 と は 、 例 え ばデ ィ ス プ レ ィ ゃプ リ ン タ 等 を 指す。 な お 、 先 の 実施例 に あ る よ う に 、 フ レ ー ム 単位 の サ イ ズ変換 と 、 小領域単位 の サ イ ズ変換 を組み合 わせ た符 号化 · 復 号化装置 の 場合 、 フ レ ー ム 単位サ イ ズ変換率の 情報 は 、 図 6 0 に示 さ れ る 小領域単位の符 号列 に先 だ っ て 再生 さ れて い な け れ ば な ら な い た め 、 フ レ ー ム 単位サ イ ズ変換率 の 符 号 は該 フ レ ー ム に お け る 全て の 小領域単位 の 符 号列 の 前 に位 置す る 。
こ の よ う に 、 蓄積媒体 に コ ン テ ン ツ と し て の 動画像 と ア ル フ ァ マ ッ プ信 号 と を 、 圧縮符号化 し て ビ ッ ト ス ト リ ー ム と し て 格納 し た場合 に 、 そ の 蓄積媒体 の 再生 シ ス テ ム を提供で き る 。
動 き べ ク ト ル の 予測符 号化 に 関す る 実施例 を 第 8 の 実施例 と し て 説 明 す る 。
図 1 9 お よ び図 2 0 は 、 本発 明 の フ レ ー ム ワ ー ク を 表す 図 で あ る 。 図 1 9 に お い て 、 動 き べ ク ト ル検 出 回 路 1 7 8 で検 出 さ れた動 き べ ク ト ル は 、 信 号線 1 0 7 を介 し て M V 符 号化 回 路 1 7 9 に供給 さ れ 、 符 号化 さ れた 後 、 V L C · 多重化 回 路 1 8 0 に供給 さ れ、 他 の 符 号化情報 と 多重化 さ れて線 3 を 介 し て 出 力 さ れ る 。
図 2 0 に お レヽ て 、 V L D ' 分離化 回 路 2 1 0 に よ つ て 、 信 号線 8 を介 し て 供給 さ れた符 号化情報 か ら 分離 さ れた動 き べ ク ト ル情報 b 1 は 、 動 き べ ク ト ル再生 回 路 2 9 0 に て 動 き べ ク ト ル信号 に 再生 さ れ る 。
本実施例 は 、 前記 の M V 符 号化 回 路 1 7 9 お よ び動 き ベ ク ト ル再生 回 路 2 9 0 に 関 わ る も の で あ る 。
一般的 に 、 動 き べ ク ト ル信号の相 関 が 隣接 ブ ロ ッ ク 間 で強 い た め 、 こ の相 関 を 除去す る た め に 予測符 号化 に よ り 動 き べ ク ト ル は符 号化 さ れ る 。
図 6 4 は 、 動 き べ ク ト ルの 予測符 号化 の 一例 を 説 明 す る た め の 図 で あ る 。
図 6 4 に お い て 、 方形 の 各枠 は マ ク ロ ブ ロ ッ ク を 示 し て お り 、 下地が ド ッ ト ノ、。 タ ー ン で示 さ れた方形枠 は 、 符 号化対象 プ ロ ッ ク で あ る 。 こ の符 号化対象 プ ロ ッ ク の動 き べ ク ト ル を M V s と す る と 、 当 該符 号化対象 ブ ロ ッ ク の M V s に 対す る 予測べ ク ト ノレ M V P s を 求 め る に は 、 当 該符 号化対象 プ ロ ッ ク の 直前 の マ ク ロ ブ 口 ッ ク ( 図 6 4 に お け る 符 号化対象 ブ ロ ッ ク の 左 隣) の ブ ロ ッ ク の 動 き べ ク ト ル M V s 1 と 、 当 該符 号化対象 ブ ロ ッ ク の 直上の マ ク ロ ブ ロ ッ ク ( 図 6 4 に お け る 符 号ィヒ対象 プ ロ ッ ク の 上 隣) の ブ ロ ッ ク の 動 き べ ク ト ノレ M V s 2 、 そ し て 、 直上 の マ ク ロ ブ ロ ッ ク の 右隣の ブ ロ ッ ク の 動 き べ ク ト ル M V s を 用 レ、 る 。
こ の よ う に 、 符 号化対象 ブ ロ ッ ク の 動 き べ ク ト ル M V s に 対す る 予測べ ク ト ル M V P s は 、 一般的 に 、 符 号化対象 プ ロ ッ ク 周 辺 の ブ ロ ッ ク の動 き べ ク ト ノレ M V s 1 、 M V s 2 、 M V s 3 を 用 い て 求 め ら れ る 。
例 え ば、 M V P s の水 平 · 垂直成分 M V P s — h お よ び M V P s V は 、 以 下 の よ う に 求 め ら れ る 。
M V P s h = Median ( M V s 1 ― h 、 M V s
2 _ h M V s 3 _ h )
M V P s v = Median ( M V s 1 ― 、 M V s
2 V 、 M V s 3 _ v )
、 ゝ _ で 、 " Med ian ( ) " と は " ( ) " 内 の値の 中 央値 を 求 め る 処理 で あ り 、 動 き ベ ク ト ル M V s n ( n
2 , 3 ;) の 水 平 · 垂直成分 を そ れぞれ、 M V s n — h 、 M V s n 一 V
と 表記 し て レヽ る 。
ま た 、 予測べ ク ト ル M V P s を 求 め る 別 の 例 と し て M V s 1 、 M V s 2 、 M V s 3 の I噴 に そ れぞれの プ ロ ッ ク に動 き べ ク 卜 ルが 存在す る カゝ否 カゝ を チ ェ ッ ク し 、 最初 に 動 き べ ク 卜 ルが 存在す る ブ ロ ッ ク の 動 き べ ク ト ル を M V P s と す る 方法 も あ る 。
図 6 5 A お よ び 6 5 B は 、 時刻 n — 1 と 時刻 n に お け る フ レ ー ム 画像 F n — 1 、 F n と そ れぞれの フ レ ー ム に お け る 符 号化領域 C A n — :! 、 C A n の例 で あ る こ こ で 、 符 号化対象 ブ ロ ッ ク 周 辺 の ブ ロ ッ ク が ォブ ジ ェ ク ト を含 ま な レ、 ブ ロ ッ ク の 場合 、 そ の ブ ロ ッ ク に は 動 き べ ク ト ル が 存在 し な い。 ま た 、 フ レ ー ム 内符 号化 さ れた ブ ロ ッ ク の場合 に も 、 そ の プ ロ ッ ク に は動 き べ ク ト ル カ S存在 し な い。
例 え ば、 M V s 1 , M V s 2 , M V s 3 の 全て が 存 在 し な い場合 、 予測べ ク ト ル M V P s に はデ フ ォ ル ト 値 (ベ ク ト ル) を用 レ、 る こ と に な る 。
オブ ジ ェ ク ト の動 き が 小 さ レヽ場合 は 、 こ の デ フ ォ ル ト 値 を零 な る 動 き ベ ク ト ル で あ る "ゼ ロ ベ ク ト ル " と し て も 問題 な い 力 s 、 図 6 5 A の フ レ ー ム 力 ら 図 6 5 B の フ レ ー ム へ の遷移 の よ う に 、 フ レ ー ム 内 で の ォブ ジ ェ ク ト の位置 の 動 き が 大 き い場合 に は 、 動 き べ ク ト ル の 予測 が 当 た ら な く な り 、 符 号化効率 が低下す る 。
本発 明 で は こ の デ フ ォ ル ト 値 と し て 、 図 6 5 A に示 さ れたべ ク ト ノレ " prev— ref " と 図 6 5 B に示 さ れた べ ク ト ノレ " curr ref " と の 差分べ ク ト ノレ " of f set" と 、 "ゼ ロ べ ク ト ル " と を 適応的 に切 り 換 え て 用 レヽ る こ と を 特徴 と す る 。
こ こ で 、 " of f set" は以 下 の 式で求 め ら れ る 。
of f set=prev ref -curr ref
デ フ オ ノレ ト イ直 " of f set" と "ゼ ロ べ ク ト ノレ " の 切 り 替 え は 、 例 え ば、 フ レ ー ム の座標軸 に基づ い て 求 め ら れた 時刻 n の オブ ジ ェ ク ト と 時刻 n — 1 の オブ ジ ェ ク ト の 誤差値 と 、 符 号化領域の座標軸 に 基づ い て 求 め ら れた 時刻 n 時点 で の オブ ジ ェ ク ト と 時刻 n — 1 時点 で の オブ ジ ェ ク ト の 誤差値 を 比較 し て 、 前者 が 大 き い 場合 は " of f set" を 、 ま た 、 後者が 大 き い場合 は "ゼ ロ ベ ク ト ル " を 、 デ フ ォ ル ト 値 と し て 用 い る よ う にすれば よ い。
こ の場合 、 切 り 替 え 情報 と し て 1 ビ ッ ト の サ イ ド情 報 を 送 る 必要 が あ る 。 ま た 、 こ の よ う な動 き べ ク ト ル の 予測ィ直の デ フ オ ル ト イ直 と し て 、 " of f set" と "ゼ 口 べ ク ト ル " を 切 り 替 え る こ と は 、 テ ク ス チ ャ 情報 の 動 き べ ク ト ル の 予測符 号化 に も 同様 に適用 で き る 。
次 に 、 図 1 9 に示 し た M V 符 号化 回 路 1 7 9 の 具体 的構成例 と 、 図 2 0 に示す M V 再生回 路 2 9 0 の 具体 的構成例 を 、 第 9 の 実施例 と し て 説明 す る 。
図 6 6 A お よ び 6 6 B は 、 図 1 9 に示 し た シ ス テ ム に お け る M V 符 号化 回 路 1 7 9 及びそ の 付 帯 回 路 の 実 施例 を 示すブ ロ ッ ク 図 で あ り 、 図 6 6 A は付 帯 回 路 と し て の デ フ オ ル ト 値演算 回 路 、 図 6 6 B は M V 符 号化 回路 1 7 9 で あ る 。
図 6 6 A に示す付帯 回 路 と し て の デ フ ォ ル ト 値演算 回 路 は 、 オブ ジ ェ ク ト 領域 に つ い て 、 現時点 で の 処理 フ レー ム と 、 前時点 で の 処理 フ レ ー ム と か ら 、 前時点 か ら 見 た現時点 で の フ レー ム の オブ ジ ェ ク ト の位置移 動 を 、 フ レ ー ム 間 で の 動 き ベ ク ト ル値 と し て 求 め る た め の も の で あ り 、 図 6 6 A に示す よ う に 、 符 号化領域 検 出 回 路 9 1 0 、 デ フ ォ ル ト 値決定回 路 9 1 1 、 領域 情報 メ モ リ 9 1 2 、 オ フ セ ッ ト 算 出 回 路 9 1 3 、 セ レ ク タ 9 1 4 に よ り 構成 さ れ る 。
ま た 、 信号線 9 0 2 は現時点 で の フ レ ー ム デー タ を 入力 す る た め の信 号線で あ り 、 図 1 9 に お け る シ ス テ ム の信 号線 2 に該 当 す る も の で 、 信 号線 2 か ら 入力 さ れ る 現時点 の フ レ ー ム デー タ を 入力 と し て 受 け る た め の も の で あ る 。 ま た 、 図 6 6 A に お け る 信 号線 9 0 2 は 、 フ レ ー ム メ モ リ 1 3 0 に保持 さ れた 前時点 フ レ ー ム デー タ を供給す る た め の信 号線で あ り 、 こ の 信 号線 9 0 2 を 介 し て 、 フ レ ー ム メ モ リ 1 3 0 力 ら 前時点 の フ レ ー ム デー タ の 供給 を 受 け る 。 ま た 、 信 号線 9 0 3 はデ フ ォ ル ト 値決定回 路 9 1 1 力ゝ ら の フ ラ グの 情報 を 出 力す る 信 号線で あ り 、 信 号線 9 0 4 は符 号化領域検 出 回路 9 1 0 か ら の符 号化領域 C A n の 情報 を 供給す る 信号線で あ り 、 信 号線 9 0 6 は領域情報 メ モ リ 9 1 2 カゝ ら 読み 出 さ れ る 符 号化領域 C A n — 1 の位置情報 を供給す る 信号線で あ る 。
ま た 、 前記符 号化領域検 出 回 路 9 1 0 は 、 信 号線 9 0 1 を介 し て 供給 さ れ る 現時点 フ レ ー ム F n - 1 の 画 像信号 を も と に 、 符 号化領域 C A n の サ イ ズお よ び位 置情報 V C 0 を 検 出す る も の で あ り 、 そ の 検 出 結果 は信号線 9 0 4 を 介 し て 、 デ フ ォ ル ト 値決定回 路 9 1 1 、 領域情報 メ モ リ 9 1 2 お よ びオ フ セ ッ ト 算 出 回 路 9 1 3 に供給す る よ う に し て あ る 。
ま た 、 領域情報 メ モ リ 9 1 2 は 、 符 号化領域 C A n - 1 の サ イ ズお よ び位置 の 情報 を 蓄積す る メ モ リ で あ り 、 時刻 n の 時点 で の フ レ ー ム の符 号化 が 終 了 後 、 符 号化領域 C A n の サ イ ズお よ び位置情報 を 蓄積す る こ と に な る 。
オ フ セ ッ ト 算 出 回 路 9 1 3 は 、 信 号線 e 4 を 介 し て 供給 さ れ る 符号化領域 C A n と 、 信 号線 9 0 6 を介 し て 供給 さ れ る 符 号化領域 C A n — 1 の位置情報 を 用 い て 、 べ ク ト ル値 " of f set" を 求 め 、 そ れ を セ レ ク タ 9 1 4 に 与 え る も の で あ る 。
ま た 、 セ レ ク タ 9 1 4 は 、 零 の 動 き ベ ク ト ル値で あ る "ゼ ロ べ ク ト ル " と オ フ セ ッ ト 算 出 回 路 9 1 3 力 ら 与 え ら れ る " of f set" と を入力 と し 、 こ れ ら の いず れか一方 を信 号線 9 0 3 を介 し て デ フ オ ル ト 値決定回 路 9 1 1 か ら 供給 さ れ る フ ラ グ に従 っ て 選択す る 回 路 で あ り 、 こ の セ レ ク タ 9 1 4 の 選択 し たべ ク ト ル値 を デ フ ォ ル ト 値 と し て 信 号線 9 0 5 を 介 し て M V 符 号化 回 路 1 7 9 の セ レ ク タ 9 2 3 に 出 力 す る 回 路 で あ る 。
以 上が 、 デ フ ォ ル ト 値演算 回 路 の構成 で あ る 。
次 に 、 M V 符 号化 回 路 1 7 9 の構成 を 説 明 す る 。 M V 符 号化 回路 1 7 9 は 、 図 6 6 B に 示す よ う に 、
M V メ モ リ 9 2 1 、 M V 予測 回 路 9 2 2 、 セ レ ク タ 9 2 3 、 差分回路 9 2 4 よ り 構成 さ れ る 。
こ れ ら の う ち 、 M V メ モ リ 9 2 1 は 図 1 9 に お け る 信 号線 1 0 7 を 介 し て供給 さ れ る 動 き べ ク ト ル検 出 回 路 1 7 8 力 ら の 動 き べ ク ト ル情報 を保存す る メ モ リ で あ り 、 符 号化対象 ブ ロ ッ ク の 周 囲 の 動 き ベ ク ト ル M V s n ( n - 1 , 2 , 3 ) カ こ の M V メ モ リ 1 7 9 に保 持 さ れ る 。
M V 予測 回 路 9 2 2 は 、 M V メ モ リ 9 2 1 か ら 供給 さ れ る 符 号化対象 ブ ロ ッ ク の 周 囲 の 動 き べ ク ト ル M V s n ( n - 1 , 2 , 3 ) を も と に 、 予測ベ ク ト ル M V P s を 求 め る 回 路 で あ る 。 こ こ で 、 M V s n ( n — 1 , 2 , 3 ) が 存在 し な い場合 は 、 予測ベ ク ト ル M V P s が 正 常 に 求 め ら れな い た め 、 当 該 M V 予測 回 路 1 Ί 9 2 は 、 予測べ ク ト ル M V P s が 正 常 に 求 め ら れた か否 か を識別す る 信号 を 出 力 す る 機能 を備 え て お り 、 こ の 識別信 号 を信号線 9 2 5 を 介 し て セ レ ク タ 9 2 3 に供 給す る 仕組み と し て あ る 。
セ レ ク タ 9 2 3 は 、 M V 予測 回 路 9 2 2 よ り 供給 さ れ る M V P s と 、 信 号線 9 0 5 を介 し て 与 え ら れ る デ フ ォ ル ト 値 と を 入力 と し 、 前記信 号線 9 2 5 を 介 し て 供給 さ れ る 信号 に従 っ て 、 両者の う ち の 一方 を 選択 し て 、 差分 回 路 9 2 4 に供給す る も の で あ る 。
差分回 路 9 2 4 は 、 動 き べ ク ト ルの 予測誤差信 号 を 求 め る 回 路 で あ っ て 、 信 号線 1 0 7 を 介 し て 供給 さ れ る 動 き べ ク ト ル検 出 回 路 1 Ί 8 力ゝ ら の 動 き べ ク ト ル情 報 と 、 セ レ ク タ 9 2 3 を介 し て 与 え ら れ る M V P s ま た はデ フ ォ ル ト 値 と の 差 を 求 め て 、 そ の結果 を 動 き べ ク ト ル情報 b 1 と し て M V 符 号化 回 路 1 7 9 カゝ ら 出 力 す る も の で あ る 。 次 に こ の よ う な構成 の 符 号化 回路 の 作用 を 説 明 す る c 図 6 6 A に お レヽ て 、 信 号線 9 0 1 に は フ レ ー ム メ モ リ 1 3 0 に保存 さ れた 前時点 で の 画像信 号 (前時点 で の フ レ ー ム の フ レ ー ム デ ー タ ) で あ る フ レ ー ム F n の 画像信号が供給 さ れ、 信 号線 e 2 に は現時点 で の 画像 信号 (現時点 で の フ レ ー ム の フ レー ム デー タ ) で あ る フ レ ー ム F n — 1 の 画像信 号が 供給 さ れ る 。
そ し て 、 フ レ ー ム F n の 画像信 号は デ フ ォ ル ト 値 決 定回 路 9 1 1 に 入力 さ れ、 フ レ ー ム F n — 1 の 画像信 号 はデ フ ォ ル ト 値決 定 回 路 9 1 1 と 符 号化領域検 出 回 路 9 1 0 に入力 さ れ る 。
符 号化領域検 出 回 路 9 1 0 で は 、 フ レ ー ム F n _ l の 画像信号 を も と に 、 符 号化領域 C A n の サ イ ズお よ び位置情報 V C 0 が 検 出 さ れ、 そ の 検 出 結果 は信 号 線 9 0 4 を介 し て 、 デ フ ォ ル ト 値決定回 路 9 1 1 、 領 域情報 メ モ リ 9 1 2 お よ びオ フ セ ッ ト 算 出 回 路 9 1 3 に供給 さ れ る 。
一方 、 デ フ ォ ル ト 値決定回 路 9 1 1 で は 、 信号線 9 0 4 を介 し て 供給 さ れ る 符 号化領域検 出 回路 9 1 0 か ら の符 号化領域 C A n の 情報 と 、 信 号線 9 0 6 を 介 し て 領域情報 メ モ リ 9 1 2 か ら 供給 さ れ る 符 号化領域 C A n - 1 の サ イ ズお よ び位置情報 を 用 い て 、 フ レ ー ム F n と フ レ ー ム F n _ l と の 誤差量 と 、 符 号化領域 C A n と C A n — 1 と の 誤差量 と の 比較 を行 う 。 そ し て 、 そ の 比較 の結果 、 前者 が 大 き い場合 に はデ フ オ ル ト 値 と し て " of f set" を 用 い 、 そ れ以外の 場合 はデ フ ォ ル ト 値 と し て ゼ ロ ベ ク ト ル を 用 レヽ る よ う に 決 定 し 、 デ フ ォ ル ト 値 と し て " of f set" を 用 レ、 る 力、 、 ゼ ロ べ ク ト ル を用 い る か、 を識別す る フ ラ グ の 情報 を信 号線 9 0 3 を介 し て 出 力 す る 。
信 号線 9 0 3 を 介 し て デ フ ォ ル ト 値決 定回 路 9 1 1 力 ら 出 力 さ れた フ ラ グの 情報 は 、 信 号線 9 0 4 を 介 し て符号化領域検 出 回 路 9 1 0 か ら 出 力 さ れた符 号化領 域 C A n の サ イ ズお よ び位置 の 情報 と 共 に 、 図 5 1 の デー タ フ ォ ーマ ツ ト に お け る 符 号化領域 の レイ ヤ に 多 重化 さ れ る 。 そ し て 、 伝送 あ る い は記録媒体へ の保存 に供 さ れ る 。
一方 、 領域情報 メ モ リ 9 1 2 は 、 符 号化領域 C A n 一 1 の サ イ ズお よ び位置情報 を 蓄積す る メ モ リ で あ り 、 時刻 n の 符号化 が 終了 後 、 符 号化領域 C A n の サ イ ズ お よ び位置情報 を 蓄積す る こ と に な る 。
オ フ セ ッ ト 算 出 回 路 9 1 3 で は 、 信 号線 9 0 4 を 介 し て供給 さ れ る 符 号化領域 C A n と 、 信 号線 9 0 6 を 介 し て 供給 さ れ る 符 号化領域 C A n — 1 の位置情報 を 用 い て 、 べ ク ト ル値 " of f set" を 求 め た 後 、 セ レ ク タ 9 1 4 に供給す る 。
セ レ ク タ 9 1 4 で は 、 信 号線 9 0 3 を 介 し て デ フ ォ ル ト 値決 定回 路 9 1 1 カゝ ら 供給 さ れ る フ ラ グ に 従 っ て 、 " of f set" と ゼ ロ べ ク ト ル と の何れか が選択 さ れ、 こ れ を デ フ ォ ル ト 値 と す る 。 こ の デ フ ォ ル ト 値 は信 号 線 9 0 5 を 介 し て M V 符 号化 回 路 1 7 9 の セ レ ク タ 9 2 3 に 出 力 さ れ る 。
図 6 6 B の M V 符 号化 回 路 1 7 9 に は 、 符 号化対象 ブ ロ ッ ク の動 き べ ク ト ル M V s が 信号線 1 0 7 を 介 し て 供給 さ れ、 M V メ モ リ 9 2 1 お よ び差分 回 路 9 2 4 に供給 さ れ る 。
M V 予測 回路 9 2 2 に は 、 M V メ モ リ 9 2 1 力ゝ ら 符 号ィヒ対象 ブ ロ ッ ク の 周 囲 の 動 き べ ク ト ノレ M V s n ( n — 1 , 2 , 3 ) ; ^ 供給 さ れ、 予測 ベ ク ト ル Μ V Ρ s 力 S 求 め ら れ る 。 こ こ で 、 M V s n ( n — 1 , 2 , 3 ) カ 存在 し な い場合 は 、 予測べ ク ト ル M V P s が 正 常 に 求 め ら れな い た め 、 予測べ ク ト ル M V P s が 正 常 に 求 め ら れた か否 か を識別す る 信 号 を 生成 し て こ れ を 信 号線 9 2 5 を 介 し 、 セ レ ク タ 9 2 3 に供給す る 。
セ レ ク タ 9 2 3 で は 、 信 号線 9 2 5 を 介 し て 供給 さ れ る 識別信 号 に従 っ て 、 M V 予測 回 路 9 2 2 よ り 供給 さ れ る M V P s 力、 、 ま た は 、 信号線 9 0 5 の 出 力 で あ る デ フ ォ ル ト 値かが 選択 さ れ、 差分回 路 9 2 4 に供給 さ れ る 。
差分回 路 9 2 4 で は 、 動 き ベ ク ト ル の 予測誤差信 号 が 計算 さ れ、 そ の 計算結果 は動 き べ ク ト ル情報 b 1 と し て M V 符 号化 回 路 1 7 9 カゝ ら 出 力 さ れ る 。
以 上 は動 き べ ク ト ル符 号化 の 処理 内 容 で あ っ た。 次 に こ の よ う に し て 符 号化 さ れた動 き べ ク ト ル の復 号化 処理 に つ い て 説 明 す る 。 図 6 7 A お よ び 6 7 B は 、 本発 明 を 実現す る た め の 復 号化 回 路 の 実施例 を 示す プ ロ ッ ク 図 で あ り 、 M V 符 号化 さ れたデー タ の 再生 を す る た め の 要部構成 を 示 し た ブ ロ ッ ク 図 で あ っ て 、 図 2 0 に示 し た シ ス テ ム に お け る M V 再生回路 2 9 0 及びそ の 付帯 回 路 の 実施例 を 示すブ ロ ッ ク 図 で あ る 。 図 6 7 A は付帯回路 と し て の デ フ ォ ル ト 値演算 回路 、 図 6 7 B は M V 再生 回 路 2 9 0 で あ る 。
図 6 7 A に示す付帯 回 路 と し て の デ フ ォ ル ト 値演算 回 路 は 、 領域情報 メ モ リ 1 0 1 0 、 オ フ セ ッ ト 算 出 回 路 1 0 1 1 、 セ レ ク タ 1 0 1 2 と 力 ら 構成 さ れ る 。 ま た 、 1 0 0 1 は符 号化 さ れて伝送 さ れて き た デー タ あ る い は記憶媒体 に記憶 さ れて 読み 出 さ れたデー タ の 上 位 レイ ヤ に含 ま れて い る デ フ ォ ル ト 値選択識別用 の フ ラ グの 情報 を 与 え る 信 号線、 1 0 0 2 は符 号化 さ れて 伝送 さ れて き た デー タ の 上位 レイ ヤ に含 ま れて い る 領 域 C A n の位置情報 を 与 え る 信 号線で あ り 、 こ れ ら は 符 号化 回 路側 に お け る 1 0 0 3 , 1 0 0 4 に 対応す る 。
ま た 、 1 0 0 3 は領域 C Λ n — 1 の位置情報 を 与 え る 信号線、 1 0 0 4 はデ フ ォ ル ト 値 を 出 力 す る 信 号線 で あ る 。
前記領域情報 メ モ リ 1 0 1 0 は 、 領域 C A n — 1 の 位置情報 を 蓄積す る メ モ リ で あ り 、 オ フ セ ッ ト 算 出 回 路 1 0 1 1 は 、 信 号線 1 0 0 2 を 介 し て供給 さ れ る 領 域 C A n の位置情報 と 、 信 号線 1 0 0 3 を 介 し て 供給 さ れ る 領域 C A n — 1 の位置情報 と を用 い て べ ク ト ル 値 " o f f set" を 求 め る も の で あ っ て 、 こ の 求 め たベ ク ト ル値 " of f set" はセ レ ク タ 1 0 1 2 に供給す る 構成 と し て あ る 。
ま た 、 セ レ ク タ 1 0 1 2 は予 め 与 え ら れた零 の 動 き べ ク ト ル値 と 、 オ フ セ ッ ト 算 出 回 路 1 0 1 1 に 与 え ら れ る べ ク ト /レ 直 " of f set" と の う ち の 一方 を 、 信 号 線 1 0 0 1 を 介 し て 与 え ら れ る デ フ ォ ル ト 値選択識別 用 の フ ラ グの 情報対応 に選択 し て 出 力 す る も の で あ り こ の セ レ ク タ 1 0 1 2 の 出 力 が デ フ ォ ル ト 値 と し て 信 号線 1 0 0 4 に 出 力 さ れ、 M V 再生回 路 2 9 0 に 与 え ら れ る 構成 で あ る 。
以 上 が復 号側 に お け る デ フ ォ ル ト 値演算 回 路 の構成 で あ る 。
次 に M V 再生回路 1 1 0 0 の構成 を 説 明 す る 。
M V 再生 回路 1 1 0 0 は 、 図 6 7 B に示す よ う に 、 カロ算 回 路 1 1 0 1 、 セ レ ク タ 1 1 0 2 、 M V 予測 回 路 1 1 0 3 、 M V メ モ リ 1 1 0 4 と 力 ら な る 。
こ れ ら の う ち 、 力!]算 回 路 1 1 0 1 は 、 複 号化対象 ブ ロ ッ ク の 動 き べ ク ト ルの 予測誤差信号 で あ る 動 き べ ク ト ル情報 b 1 と セ レ ク タ 1 1 0 2 を介 し て 与 え ら れ る デ フ ォ ル ト 値 と を 受 け て 、 両者 を加算 し て 出 力 す る 回 路 で あ り 、 こ の力 B算 出 力 は M V メ モ リ 1 1 0 4 並び に 図 2 0 の構成 に お け る 信 号線 2 0 3 に 出 力 さ れ る 構成 で あ る 。 ま た 、 M V メ モ リ 1 1 0 4 は 、 前記加算 回 路 1 1 1 0 の加算 出 力 を保持 し て 複 号化対象 プ ロ ッ ク の 周 囲 の 動 き ベ ク ト ル M V s n ( n = l , 2 , 3 ) を 供給す る た め の も の で あ り 、 ま た 、 M V 予測 回 路 1 1 0 3 は 、 M V メ モ リ 1 1 0 4 か ら 供給 さ れ る 複 号化対象 ブ ロ ッ ク の 周 囲 の 動 き べ ク ト ノレ M V s n ( n = 1 , 2 , 3 ) か ら 予測べ ク ト ル M V P s を 求 め 、 セ レ ク タ 1 1 0 2 に 与 え る も の で あ る 。 な お 、 M V s n ( n = 1 , 2 , 3 ) が 存在 し な い場合 は 、 予測ベ ク ト ル M V P s が 正 常 に求 め ら れな い た め 、 M V 予測 回路 1 1 0 3 に は 、 予測べ ク ト ル M V P s が 正 常 に求 め ら れた か否 力ゝ を識 別す る 信号 を発 生す る 機能 を持たせ て あ り 、 こ の識別 信 号 は 、 信 号線 1 1 0 5 を 介 し て セ レ ク タ 1 1 0 2 に 供給す る 構成 と し て あ る 。
セ レ ク タ 1 1 0 2 は 、 信 号線 1 0 0 4 を 介 し て 与 え ら れ る デ フ ォ ル ト 値 と 、 M V 予測 回 路 1 1 0 3 力ゝ ら 与 え ら れ る 予測べ ク ト ル M V P s を 入力 と し 、 こ れ ら の 一方 を 、 信 号線 1 1 0 5 を 介 し て 供給 さ れ る 識別信 号 に従 っ て 、 選択 し て加算 回 路 1 1 0 1 に 与 え る 回 路 で あ る 。
次 に 、 こ の よ う な構成 の 復 号化側 シ ス テ ム の 作用 を 説 明 す る 。
図 6 7 A に お い て 、 信 号線 1 0 0 1 に はデ フ ォ ル ト 値 と し て " of f set" を 用 レヽ る 力、 、 ゼ ロ べ ク ト ル を 用 い る カゝ を識別す る フ ラ グが 供給 さ れ、 信 号線 1 1 0 2 に は領域 C A n の位置情報 が 供給 さ れ る 。
信 号線 1 0 0 2 を 介 し て 供給 さ れた領域 C A n の位 置情報 は 、 領域情報 メ モ リ 1 0 1 0 と オ フ セ ッ ト 算 出 回 路 1 0 1 1 に供給 さ れ る 。 領域情報 メ モ リ 1 0 1 0 は 、 領域 C A n — 1 の位置情報 を 蓄積す る メ モ リ で あ り 、 時刻 n の複 号化 が 終了 後 、 領域 C A n の位置情報 を 蓄積す る こ と に な る 。
オ フ セ ッ ト 算 出 回 路 1 0 1 1 で は 、 信 号線 d 2 を 介 し て 供給 さ れ る 領域 C A n の位置情報 と 、 信 号線 1 0 0 3 を 介 し て 供給 さ れ る 領域 C A n — 1 の位置情報 と を 用 い て ベ タ ト ルィ直 " of f set" を 求 め た 後 、 べ ク ト ル値 " of f set" を セ レ ク タ 1 0 1 2 に供給す る 。
セ レ ク タ 1 0 1 2 で は 、 信 号線 1 0 0 1 を 介 し て 供 給 さ れ る フ ラ グ に従 っ て 、 " of f set" と "ゼ ロ べ ク ト ル " の う ち の何れか一方が 選択 さ れ、 デ フ ォ ル ト 値 と し て 信 号線 1 0 0 4 を 介 し て M V 再生回 路 1 1 0 0 へ と 出 力 す る 。
次 に 図 6 7 B の M V 再生 回 路 1 1 0 0 に お い て は 、 複 号化対象 プ ロ ッ ク の 動 き べ ク ト ルの 予測誤差信 号で あ る "動 き べ ク ト ル情報 b 1 " が 供給 さ れ、 加算 回 路 1 1 0 1 に供給 さ れ る 。
ま た 、 M V 予測 回 路 1 1 0 3 に は 、 M V メ モ リ 1 1 0 4 か ら 複 号化対象 プ ロ ッ ク の周 囲 の 動 き べ ク ト ル M V s n ( n = 1 , 2 , 3 ) 力 S 供給 さ れ、 予測ベ ク ト ル M V P s 力 求 め ら れ る 。 こ こ で 、 M V s n ( n = 1 , 2 , 3 ) が存在 し な い場合 は 、 予測べ ク ト ル Μ V P s が 正 常 に 求 め ら れ な い た め 、 予測べ ク ト ル Μ V P s が 正 常 に求 め ら れた か否か を識別す る 信 号 を信 号線 1 1 0 5 を介 し て セ レ ク タ 1 1 0 2 に供給す る 。
セ レ ク タ 1 1 0 2 で は 、 信 号線 1 1 0 5 を 介 し て 供 給 さ れ る 当 該識別信号 に従 っ て 、 Μ V 予測 回 路 1 1 0 3 よ り 供給 さ れ る Μ V P s カゝ 、 ま た は信 号線 1 0 0 4 の 出 力 で あ る デ フ オ ル ト 値の レヽずれかが 選択 さ れ、 力 Π 算 回 路 1 1 0 1 に供給 さ れ る 。
加算 回 路 1 1 0 1 で は 、 動 き べ ク ト ル の 予測誤差信 号 ( "動 き べ ク ト ル情報 b 1 " ) と 予測信 号 M V P s と が加算 さ れ、 こ れ に よ つ て 複 号化対象 ブ ロ ッ ク の動 き ベ ク ト ル M V s が 再生 さ れ る 。 こ の 復 号化対象 ブ ロ ッ ク の動 き べ ク ト ル M V s は信 号線 2 0 3 を 介 し て M V 再生 回路 1 1 0 0 カゝ ら 出 力 さ れ る と 共 に 、 M V メ モ リ 1 1 0 4 に 蓄積 さ れ る 。
こ の よ う に し て 、 図 1 9 に示す構成 で の 必要 な M V 符 号化処理 と 、 図 2 0 に示す構成 で の 必要 な M V 再生 処理が実現で き る 。 産業 上 の利用 可能性
以 上 、 種 々 の 実施例 を 説 明 し た が 、 本発 明 に よ れ ば、 オブ ジ ェ ク ト の 形状や画面 内 の位置 な ど を 表す副 画像 情報 で あ る ア ル フ ァ マ ッ プ の 情報 を 効率 よ く 符 号化 で き る と 共 に 、 そ の 復 号 を 行 う こ と が で き る よ う に し た 画像符 号化装置 お よ び復 号化装置 が得 ら れ る 。
ま た 、 本発 明 に よ れば、 ア ル フ ァ マ ッ プ の符 号量 を 低減す る こ と が で き る た め 、 フ レ ー ム 単位 で符 号化す る 従来の符 号化法 と 比べ て 、 大幅 な符 号化効率 の 低下 な し に オブ ジ ェ ク ト 毎 に別 々 に符 号化す る こ と が で き る よ う に な る 。
な お 、 本発 明 は上述 し た 実施例 に 限定す る こ と な く 種 々 変形 し て 実施 し得 る 。
本発 明 に よ れば、 ア ル フ ァ マ ッ プの符 号量 を 低減す る こ と が で き る た め 、 フ レ ー ム 単位 で符 号化す る 従来 の 符号化法 と 比べ て 、 大幅 な 符 号化効率 の 低下 な し に オ ブ ジ ェ ク ト 毎 に別 々 に符 号化す る こ と が で き る よ う に な る 。

Claims

請 求 の 範 囲
1 . オブ ジ ェ ク ト の 形状お よ びそ の オブ ジ ェ ク ト の 画面 内 で の位置 を 表すア ル フ ァ マ ッ プ信号 を 、 サ イ ズ 変換率 ( s ize convers ion ratio )情報 に従 っ た縮 小率で ダ ウ ン サ ン プルす る 第 1 の ダ ウ ン サ ン プ リ ン グ
( down-sampling ) ID と 、
前記第 1 の ダ ウ ン サ ン プ リ ン グ回 路 に よ っ て ダ ゥ ンサ ン プル さ れた ア ル フ ァ マ ッ プ信 号 を 元 の サ イ ズ に 戻すた め 与 え ら れ る サ イ ズ変換率情報 に従 っ た 拡大 率 で ア ル フ ァ マ ッ プ信 号 を ア ッ プサ ン プル ( u p - sample ) し 、 ア ル フ ァ マ ッ プ局部復 号信 号 を 出 力 す る ア ッ プサ ン プ リ ン グ回 路 と 、
先 の 再生画像信 号 と 動 き べ ク ト ル信 号 と に 基づ い て 動 き 補償信号 を 生成す る 動 き 補償 回 路 と 、
前記動 き 補償信 号 を 前記縮小率 で ダ ウ ン サ ン プ ルす る 第 2 の ダ ウ ン サ ン プ リ ン グ回 路 と 、
前記第 1 の ダ ウ ンサ ン プ リ ン グ回 路 に よ っ て ダ ゥ ン サ ン プル さ れた ア ル フ ァ マ ッ プ信 号 を 前記第 2 の ダ ウ ン サ ン プ リ ン グ回 路 の ダ ウ ン サ ン プル動 き 補償予 測信 号 に従 っ て 2 値画像 に符 号化 し 、 2 値画像符 号化 信号 を 出 力 す る 2 値画像符 号化 回 路 と 、
前記 2 値画像符 号化信 号 と サ イ ズ変換率 の 情報 と を 多重化 し て 出 力 す る 多重化 回 路 と 、
で構成 さ れ る ア ル フ ァ マ ッ プ符 号化装置。
2 . 前記動 き 補償予測 回 路 は 、 先 の 再生画像信号 ( previous decoded ideo signal ) を 格納す る メ モ リ を 有 し 、 こ の メ モ リ に格納 さ れた 前記再生画像信 号 と 動 き べ ク ト ル信 号 に従 っ て 動 き 補償予測信号 を 生 成す る 請求項 1 の ア ル フ ァ マ ッ プ符 号化装置。
3 . 前記第 1 ダ ウ ン サ ン プ リ ン グ回 路 は 、 前記 ア ル フ ァ マ ッ プ信 号 を M X N 画素 ( M : 水 平方 向 の 画素数 、 N : 垂直方 向 の 画素数) の ブ ロ ッ ク 単位 で受 け 、 前記 ア ル フ ァ マ ッ プ信 号 を ブ ロ ッ ク 単位 で符 号化す る 請求 項 1 の ア ル フ ァ マ ッ プ符 号化装置。
4 . 符 号化 ア ル フ ァ マ ッ プ信 号カゝ ら ア ル フ ァ マ ッ プ 符号 と サイ ズ変換率情報符 号 と を 分離す る 分離回 路 と 、 前記ア ル フ ァ マ ッ プ符 号 を 復号 し 、 2 値画像 に 戻す 2 値画像複 号化 回 路 と 、
前記 2 値画像複 号化 回 路 か ら の ア ル フ ァ マ ッ プ 符号で あ る 2 値画像 を 、 前記サ イ ズ変換率情報符 号 に 従 っ て ア ッ プサ ン プル し て 再生画像信 号 と し て 出 力 す る ア ッ プサ ン プ リ ン グ回 路 と 、
先 に復 号化 さ れた 画像信 号 と 動 き べ ク ト ル信号 と 基づ い て 動 き 補償信 号 を 生成す る 動 き 補償 回 路 と 、
前記動 き 補償信 号 を 、 前記サ イ ズ変換率情報 に従 つ て ダ ウ ンサ ン プル し 、 ダ ウ ンサ ン プル さ れた 動 き 補 償信号 を 前記 2 値画像復 号化 回 路 に 出 力 す る ダ ウ ンサ ン プ リ ン グ回 路 と 、
に よ り 構成 さ れ、
前記 2 値画像復 号化 回 路 は前記動 き 補償予測信 号 と 前記サ イ ズ変換率情報 に従 っ て 前記ア ル フ ァ マ ッ プ 符 号 を復 号す る 、
ア ル フ ァ マ ッ プ復号化装匱。
5 . オブ ジ ェ ク ト の形状 を 表すア ル フ ァ マ ッ プ信 号 を M X N 画素 ( M : 水 平方 向 の 画素数 、 N : 垂 直方 向 の 画素数) の プ ロ ッ ク 単位 で分割 し て 得 ら れ る 複数 の ア ル フ ァ マ ッ プ信号プ ロ ッ ク を 受 け る 入力 部 と 、
前記ア ル フ ァ マ ッ プ信号 に つ い て 、 ブ ロ ッ ク 近 傍の 再生信号 ( decoded signal )を含 む フ レ ー ム の 再生信 号 と 、 符 号化 が 終了 し た フ レ ー ム の 再生信 号 を 蓄 え る フ レ ー ム メ モ リ と 、
フ レ ー ム メ モ リ 内 のす で に符 号化 が 終了 し た フ レ ー ム の 再生信号 を 用 い て 、 動 き 補償予測値 を 生成す る 動 き 補償予測手段 と 、
プ ロ ッ ク 毎 に 2 値画像 を サ イ ズ変換す る 手段 と 、 サ イ ズ変 率 ( size conversion ratio )を サ イ ド情報 と し て符 号化す る 手段 と 、
プ ロ ッ ク 毎 に縮小 さ れた 2 値画像 を 符 号化す る 2 値画像符 号化手段 と 、
に よ り 構成 さ れ る ア ル フ ァ マ ッ プ符 号化器 ( alpha-map encoder ) ¾r 有 し 、
前記ア ル フ ァ マ ッ プ符 号化器 は前記プ ロ ッ ク の 再生画像 を 、 プ ロ ッ ク 内 を 2 値の 何れか一方 の値 に全 て 置 き 換 え た再生値 と 、 動 き 補償予測値 と 、 ブ ロ ッ ク 毎 に サ イ ズ変換す る こ と に よ り 得 ら れた 再生値 と の う ち の 何れかか ら 選択す る 、 画像符 号化装置。
6 . ォ ブ ジ ェ ク ト を 含 む方形領域 を 、 M X N 画素 ( M : 水 平方 向 の 画素数、 N : 垂直方 向 の 画素数) の 単位 で分割す る こ と に よ っ て 得 ら れ る 複数 の 画像 プ ロ ッ ク を 順次受 け る 画像 プ ロ ッ ク 入力 部 と 、
ブ ロ ッ ク 近傍 の 再生信 号 を 含 む該 フ レ ー ム の 再 生信 号 と 符号化終 了 済 み フ レ ー ム の 再生信号 を 蓄 え る フ レ ー ム メ モ リ と 、
フ レ ー ム メ モ リ 内 の 符 号化終了 済 み フ レ ー ム の 再生信 号 を用 い て 、 動 き 補償予測値 を 生成す る 動 き 補 償予測手段 と 、
プ ロ ッ ク 毎 に 2 値画像 を サ イ ズ変換す る 手段 と 、 サ イ ズ変換率 を サイ ド情報 と し て 符 号化す る 手 段 と 、
ブ ロ ッ ク 毎 に縮小 さ れた 2 値画像 を 符 号化す る 2 値画像符 号化手段 と 、
2 値画像符 号化 の 符 号化順序 を ブ ロ ッ ク 内 で変 更す る 手段 と 、
力 ら な る ア ル フ ァ マ ッ プ符 号化器 を備 え 、 前記ア ル フ ァ マ ッ プ符 号化器 は前記 プ ロ ッ ク の 再生画像 を 、 ブ ロ ッ ク 内 を 2 値 の 何れか一方 の値で全 て 置 き 換 え た再生値 と 、 動 き 補償予測値 と 、 ブ ロ ッ ク 毎 に サ イ ズ変換す る 事 で得 ら れた 再生値の う ち の 何れ かか ら 選択す る 、 画像符 号化装置。
7 . 前記 2 値画像符 号化手段 は 、 切 り 換 え 情報 に応 じ て複数種 の 符 号化手法 の 1 つ を 選択 し 、 前記切 り 換 え 情報 を符 号化 し て 送 る 請求項 6 の 画像符 号化装置。
8 . ア ル フ ァ マ ッ プ符 号化手段 はオブ ジ ェ ク ト を含 む方形領域 を サイ ズ変換す る 手段 と 、 サ イ ズ変換率 を サ イ ド情報 と し て符 号化す る 手段 と を 有 し 、 前記縮小 さ れた方形領域 を 符号化対象 と す る 請求項 6 の 画像符 号化装置。
9 . ブ ロ ッ ク 近傍 の 再生信号 を含 む フ レ ー ム の 再生 信 号 と 、 複 号化済 み フ レ ー ム の 再生信 号 を 蓄 え る フ レ — ム メ モ リ と 、
フ レ ー ム メ モ リ 内 の す で に複 号化 が 終了 し た フ レ ー ム の 再生信 号 を用 い て 、 動 き 補償値 を 生成す る 動 き 補償手段 と 、
プ ロ ッ ク 毎 に 2 値画像 を サ イ ズ変換す る 手段 と 、 ブ ロ ッ ク 毎 に縮小 さ れた 2 値画像 を 複 号化す る
2 値画像複号化 手段 と か ら な る ア ル フ ァ マ ッ プ復 号化 器 を 有 し 、
前記ア ル フ ァ マ ッ プ複 号化器 は 、 前記 ブ ロ ッ ク の 再生画像 を 、 プ ロ ッ ク 内 を 2 値の何れか一方 の値で 全て 置 き 換 え た 再生値 と 、 動 き 補償予測値 と 、 前記縮 小 さ れた 2 値画像 を 拡大す る こ と で得 ら れた 再生値 の う ち の何れかか ら 選択す る 、
画像復 号化装置。
1 0 . プ ロ ッ ク 近傍 の 再生信号 を含 む該 フ レ ー ム の 再 生信 号 と 、 す で に複 号化 が 終了 し た フ レ ー ム の 再生信 号 を 蓄 え る フ レ ー ム メ モ リ と 、
フ レ ー ム メ モ リ 内 の すで に復 号化 が 終了 し た フ レ ー ム の 再生信号 を 用 い て 、 動 き 補償値 を 生成す る 動 き 補償手段 と 、
ブ ロ ッ ク 毎 に 2 値画像 を サ イ ズ変換す る 手段 と 、 プ ロ ッ ク 毎 に縮小 さ れた 2 値画像 を 復 号化す る 2 値面像複 号化手段 と 、
2 値画像復 号化 の複 号化順序 を プ ロ ッ ク 内 で変 更す る 手段 と カゝ ら な る ア ル フ ァ マ ッ プ複 号化器 と 、
を 有 し 、
前記 ア ル フ ァ マ ッ プ復 号化器 は 、 前記 ブ ロ ッ ク の 再生画像 を 、 ブ ロ ッ ク 内 を 2 値の 何れか一方 の値で 全て 置 き 換 え た再生値 と 、 動 き 補償予測値 と 、 プ ロ ッ ク 毎 に サイ ズ変換す る 事 で得 ら れた再生値の う ち の 何 れか か ら 選択す る 、 画像復 号化装置。
1 1 . 前記 ア ル フ ァ マ ッ プ複 号化手段 は 、 複 号化手法 の 異 な る 複数種 の 2 値画像複 号化手段 を 有 し 、 こ れ ら を 切 り 換 え 情報 に よ り 選択切 り 換 え し て使用 す る 請求 項 1 0 の 画像復号化装置。
1 2 . 前記 ア ル フ ァ マ ッ プ複 号化手段 は 、 オブ ジ ェ ク ト を含 む方形領域 を 拡大す る 手段 を備 え 、 縮小 さ れた 方形領域 を 拡大 し て 復 号化す る 請求項 1 0 ま た は 1 1 の 画像復 号化装置。
1 3 · オブ ジ ェ ク ト を含 む 、 ブ ロ ッ ク サ イ ズの倍数で 表 さ れ る 領域 を 設 定す る 手段 と 、 前記領域 を複数の プ ロ ッ ク に 分割す る 手段 と を 有 し 、 各 ブ ロ ッ ク に 対 し て 、 各 々 の 属性 に 固 有 の ラ ベ ル を 割 り 当 て る ラ ベル付 け 手段 と 、
前記 ラ ベル情報 と 領域の サ イ ズ を フ レ ー ム 毎 に 保持す る メ モ リ と 、
前記 メ モ リ に 蓄積 さ れた ラ ベ ル情報 を 、 現 フ レ ー ム の領域 の サ イ ズ に合わせ て 変 更す る サ イ ズ変 更手 段 と 、
該 ブ ロ ッ ク 近傍 の 再生済 み の プ ロ ッ ク の ラ ベノレ 情報 を保持す る メ モ リ と を備 え 、
該ブ ロ ッ ク の ラ ベル情報 を 、 前記サイ ズ変 更 手 段 よ り 供給 さ れ る ラ ベル情報 と 該 ブ ロ ッ ク 近傍 の プ ロ ッ ク の ラ ベル情報 の いずれか を 用 レ、 て 符 号化す る 、 画 像符 号化装置。
1 4 . ブ ロ ッ ク 毎 に ァ ノレ フ ァ マ ッ プ信 号 を プ ロ ッ ク 毎 の 属性 と 共 に符 号化 し て 得 ら れ る 符 号化デー タ を 受 け る 入力 部 と 、
前記符 号化デー タ の ラ ベル情報 と 領域の サ イ ズ を フ レ ー ム 毎 に保持す る メ モ リ と 、
前記 メ モ リ に 蓄積 さ れて い る ラ ベ ル情報 を 、 現 フ レ ー ム の領域 の サイ ズ に 合 わせ て 変 更す る サ イ ズ変 更 手段 と 、
前記ブ ロ ッ ク 近傍の 再生済 み の ブ ロ ッ ク の ラ ベ ル情報 を保持す る メ モ リ と を備 え 、
現 フ レ ー ム の ラ ベ ル情報 を 、 前記サ イ ズ変 更 手 6 9 段 よ り 供給 さ れ る ラ ベル情報 と 該 プ ロ ッ ク 近傍 の ブ ロ ッ ク の ラ ベル情報の いずれか を 用 い て 復 号化す る 画像 複 号化装置
1 5 . 画像 を 、 そ の 画像 の オブ ジ ェ ク ト 領域 と 背景領 域 に 区別す る た め の 情報 で あ る ア ル フ ァ マ ッ プ信 号 を 受 け る 入力 部 と
ォブ ジ ェ ク ト を含 む 、 プ π ッ ク サ イ ズの倍数 で 表 さ れ る 符号化領域 を 設 定す る 手段 と 、
HU 記領域 を複数 の ブ ロ ッ ク に 分割す る 手段 と 、 各 ブ Π -y ク に対 し て 、 各 々 の 属性 に 固 有 の ラ ベ ル を 割 り 当 て る ラ ベル付 け 手段 と 、
nn 記 ラ ベル あ る い は前記ァ ル フ ァ マ ッ プ信 号 を フ レ ー ム 毎 に保持す る 記億手段 と 、
現 フ レ一ム の 符 号化 プ ロ ッ ク に対応す る 前 フ レ ー ム の 参 照 ブ 口 ッ ク を 決定す る 決定手段 と 、
少 な < と も 前記記憶手段 に保持 さ れた 前 フ レ ー ム の ラ ベ /レ あ る い は ァ ノレ フ ァ マ ッ プ と 、 前記参 照 プ ロ ッ ク に よ つ て 予測値 を 決定す る 予測手段 と 、
刖 目己符 号化 プ ロ ッ ク の ラ ベ ノレ情報 を 、 前記予測 値 を用 い て符号化す る 符 号化手段 と
に よ り 構成 さ れ る 画像符 号化装置。
1 6 . プ ロ ッ ク 近傍 の 再生信 号 を含 む フ レ ー ム の 再生 信 号 を 蓄 え る フ レ ー ム メ モ リ と
ブ 口 ッ ク 毎 に 2 値画像 を サ イ ズ変換す る 手段 と プ 口 ッ ク 毎 に縮小 さ れた 2 値画像 を 復 号化す る 2 値画像復 号化手段 と カゝ ら な る ア ル フ ァ マ ッ プ復 号化 器 を 有 し 、
ブ ロ ッ ク の 拡大 に お い て は 、 既 に符号化 し た ブ ロ ッ ク の 再生値 を 参 照 し な が ら 拡大す る 画像復 号化装 置。
1 7 . オブ ジ ェ ク ト を含 む M X N 画素 ( M : 水 平方 向 の 画素数 、 N : 垂直方 向 の 画素数) の 符 号化画像 プ ロ ッ ク を 入力 と す る 入力 部 と 、
ブ ロ ッ ク 近傍 の 再生信 号 を 含 む フ レ ー ム の 再生 信 号 を 蓄 え る フ レ ー ム メ モ リ と 、
ブ ロ ッ ク 毎 に 2 値画像 を サ イ ズ変換す る 手段 と 、 プ ロ ッ ク 毎 に縮小 さ れた 2 値画像 を 複 号化す る 2 値画像復 号化手段 と か ら な る ア ル フ ァ マ ッ プ複 号化 器 を ¾ し 、
前記ア ル フ ァ マ ッ プ複 号化器は 、 ブ ロ ッ ク の 拡 大処理時 、 既 に符 号化 し た プ ロ ッ ク の 再生値 を 参 照 し て 処理す る か 、 若 し く は参 照 し な い で処理す る かいず れか を 切 り 替 え て 処理実行す る 機能 を備 え る 画像復 号 化装置。
1 8 . オブ ジ ェ ク ト を含 む方形領域 を 、 M X N 画素
( M : 水 平方 向 の 画素数 、 N : 垂直方 向 の 画素数) の 単位 で分割す る こ と に よ っ て 得 ら れ る 複数の 画像 プ ロ ッ ク を順次受 け る 画像 プ ロ ッ ク 入力 部 と 、
オ ブ ジ ェ ク ト の 形状 を表す 2 値画像 で あ る ア ル フ ァ マ ッ プ信 号 に つ い て 、 前記各 ブ ロ ッ ク を そ の ブ ロ ッ ク の 水平 · 垂直方 向 共 に l Z 2 n ( n = 0 , 1 2 3 , ··· ) に縮小処理す る 手段 と 、
刖記 is小処理 さ れた プ ロ ッ ク を 拡大処理す る 拡 大 回 路 と を 有 し 、 符 号化 に あ た っ て は レ ー ト 制御 の た め に n の値 を 送 る 手段 と 、
を 具備 し
刖 記拡大 回 路 は 、
前記ブ ロ ッ ク の 近傍再生値 を保持す る メ モ リ と 該 ブ 口 ッ ク の 縮小率 に応 じ て 前記 メ モ リ に保持 さ れた 再生値 を 1 Z 2 n に縮小す る こ と で参 照 画素 値 を 求 め る 手段 と 、
水 平 · 垂直方 向 共 に 2 倍 に 拡大す る 処理 を n 回 繰 り 返す こ と で元 の サ イ ズ に拡大す る 拡大処理手段 と を備 え 、
前記拡大処理手段 に お い て は 、 常 に 1 Z 2 n に縮小 さ れた 参 照 画素値 を 用 い る 画像符 号化装置。 1 9 . 水 平 · 垂直共 に l / 2 n ( n = l , 2 , 3 ,
… ) に縮小 さ れた 2 値画像 の プ ロ ッ ク を 拡大す る 拡大 回路 で め っ て ゝ
該ブ 口 ッ ク 近傍再生値 を保持す る メ モ リ と 、 該 ブ 口 ッ ク の 縮小率 に応 じ て 前記 メ モ リ に保持 さ れた再生値 を 1 / 2 n に縮小す る こ と で参 照 画素 値 を 求 め る 手段 と 、
水 平 ' 垂直共 に 2 倍 に 拡大す る 処理 を n 回繰 り 返す こ と で元 の サ イ ズ に 拡大す る と 共 に 、 こ の 拡大処理 に 2 お い て は 、 常 に 1 Z 2 n に縮小 さ れた 参 照 画素値 を 用 い る 拡大処理手段 と 、
を 具備す る 画像複 号化装置。
2 0 . オブ ジ ェ ク ト を 含む方形領域 を 、 M X N 画素 ( M : 水平方 向 の 画素数 、 N : 垂直方 向 の 画素数) の 単位で分割す る こ と に よ っ て 得 ら れ る 複数の 画像 プ ロ ッ ク を 順次受 け る 画像 プ ロ ッ ク 入力 部 と 、
前記 フ レ ー ム 内 に お い て 、 オブ ジ ェ ク ト を含 む プ ロ ッ ク サ イ ズの倍数で表 さ れ る 領域 を設 定す る 設 定 手段 と 、
刖 目 ΰ設 定手段 に よ り 設 定 さ れた領域 内 を プ ロ ッ ク に分割す る 分割手段 と 、
分割 さ れた 前記 プ ロ ッ ク 内 を 動 き 補償予測す る た め に必要 な 動 き べ ク ト ル を 予測符 号化す る 手段 と 、 参 照 フ レ一ム 内 の領域の フ レ ー ム 内 に お け る 位 置 を 表す第 1 の位置べ ク ト ル を保持す る メ モ リ と 、 該 フ レ一ム 内 領域 の フ レ ー ム 内 で の位置 を 表す 第 2 の位置べ ク ト ル を符 号化す る 符 号化手段 と 、
符 号化対象 プ ロ ッ ク 近傍 の 再生済みの プ ロ ッ ク の動 き べ ク ト ル を保持す る 動 き べ ク ト ル メ モ リ と 、 m 記動 き べ ク ト ル メ モ リ に 蓄 え ら れて レ、 る 動 き べ ク ト ル を用 い て 、 符 号化対象 ブ ロ ッ ク の動 き べ ク ト ル を 予測す る 手段 と 、
を 具備 し 、
m 記予測手段 に お い て使用 す る 動 き べ ク ト ルが 前記動 き ク ト ノレ メ モ リ に存在 し な レヽ場合 に はデ フ ォ ル ト の 動 き べ ク ト ル を 予測値 と し 、 こ の デ フ ォ ル ト の 動 き べ ク 卜 ル は 第 1 の位置べ ク ト ル と 第 2 の位置べ ク ト ル と の 差分べ ク ト ノレ と 、 ゼ ロ べ ク ト ル を切 り 換 え て 用 い る 画像符号化装置
2 1 . ォプ ジ ェ ク ト を含 む Μ X Ν 画素 ( Μ : 水 平方 向 の 画素数 、 Ν : 垂直方 向 の 画素数) の符 号化画像 プ ロ ッ ク を入力 と す る 入力 部 と 、
、 _ 一
刖 己 ブ ロ ッ ク 内 を 動 き 補償予測す る た め に 必要 な 、 予測符 号化 さ れた動 き べ ク ト ル を復 号す る 手段 と 参照 フ レ 一 ム 内 を 動 き 補償予測す る た め に 必要 な 、 予測符号化 さ れた動 き べ ク ト ル を 復 号す る 手段 と 参照 フ レ ー ム 内 の領域 の フ レ ー ム 内 で の位置 を 表す第 1 の位置べ ク ト ル を保持す る メ モ リ と 、
該 フ レ ー ム 内 の領域 の フ レ ー ム 内 で の位置 を 表 す第 2 の位置べ ク ト ノレ を復 号化す る 手段 と 、
復 号化対象 ブ ロ ッ ク 近傍 の補正済 みの プ ロ ッ ク の動 き べ ク ト ル を保持す る 動 き べ ク ト /レ メ モ リ と 、 刖 己動 き べ ク 卜 ノレ メ モ リ に保持 さ れた 動 き べ ク ト ル を 用 い て 、 復 号化対象 プ ロ ッ ク の動 き べ ク ト /レ を 予測す る 予測手段 と を 有 し 、
m 記予測手段 で使用 す る た め の 動 き べ ク ト ルが 刖 sti動 さ ク ト ノレ メ モ リ に存在 し な い場合 に はデ フ ォ ノレ ト の動 き べ ク ト ル を 予測値 と し 、 こ の デ フ ォ ル ト の 動 き べ ク 卜 ルは第 1 の位置べ ク ト ル と 第 2 の位置べ ク ト ノレ と の差分べ ク ト ル と 、 ゼ ロ べ ク ト ノレ の レ、ずれ力 と す る 画像復 号化装置。
PCT/JP1997/003976 1996-10-31 1997-10-31 Image encoder and image decoder WO1998019462A1 (en)

Priority Applications (12)

Application Number Priority Date Filing Date Title
KR1019980704912A KR100286443B1 (ko) 1996-10-31 1997-10-31 화상부호화장치 및 화상복호화장치
AU47266/97A AU713780B2 (en) 1996-10-31 1997-10-31 Video encoding apparatus and video decoding apparatus
CA 2240132 CA2240132C (en) 1996-10-31 1997-10-31 Video encoding apparatus and video decoding apparatus
EP97909712A EP0873017A4 (en) 1996-10-31 1997-10-31 IMAGE ENCODER AND DECODER
US09/091,362 US6122318A (en) 1996-10-31 1997-10-31 Video encoding apparatus and video decoding apparatus
BR9706905A BR9706905A (pt) 1996-10-31 1997-10-31 Aparelho para codificação de vídeo e aparelho para decodificação de vídeo
NO983043A NO983043D0 (no) 1996-10-31 1998-06-30 Videokode- og videodekodeapparat
US09/634,745 US6259738B1 (en) 1996-10-31 2000-08-08 Video encoding apparatus and video decoding apparatus
US10/703,667 US7308031B2 (en) 1996-10-31 2003-11-10 Video encoding apparatus and video decoding apparatus
US10/971,284 US7215709B2 (en) 1996-10-31 2004-10-25 Video encoding apparatus and video decoding apparatus
US10/971,309 US7167521B2 (en) 1996-10-31 2004-10-25 Video encoding apparatus and video decoding apparatus
US11/683,336 US20070147514A1 (en) 1996-10-31 2007-03-07 Video encoding apparatus and video decoding apparatus

Applications Claiming Priority (10)

Application Number Priority Date Filing Date Title
JP29003396 1996-10-31
JP8/290033 1996-10-31
JP9243297 1997-04-10
JP9/92432 1997-04-10
JP9/116157 1997-04-18
JP11615797 1997-04-18
JP14423997 1997-06-02
JP9/144239 1997-06-02
JP17777397A JP4034380B2 (ja) 1996-10-31 1997-06-18 画像符号化/復号化方法及び装置
JP9/177773 1997-06-18

Related Child Applications (6)

Application Number Title Priority Date Filing Date
US09091362 A-371-Of-International 1997-10-31
US09/091,362 A-371-Of-International US6122318A (en) 1996-10-31 1997-10-31 Video encoding apparatus and video decoding apparatus
US09/543,775 Division US6198768B1 (en) 1996-10-31 2000-04-05 Video encoding apparatus and video decoding apparatus
US09/634,761 Division US6292514B1 (en) 1996-10-31 2000-08-08 Video encoding apparatus and video decoding apparatus
US09/634,745 Division US6259738B1 (en) 1996-10-31 2000-08-08 Video encoding apparatus and video decoding apparatus
US09/634,550 Division US6754269B1 (en) 1996-10-31 2000-08-08 Video encoding apparatus and video decoding apparatus

Publications (1)

Publication Number Publication Date
WO1998019462A1 true WO1998019462A1 (en) 1998-05-07

Family

ID=31721799

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP1997/003976 WO1998019462A1 (en) 1996-10-31 1997-10-31 Image encoder and image decoder

Country Status (11)

Country Link
US (8) US6122318A (ja)
EP (1) EP0873017A4 (ja)
JP (1) JP4034380B2 (ja)
KR (1) KR100286443B1 (ja)
CN (2) CN100382603C (ja)
AU (1) AU713780B2 (ja)
BR (1) BR9706905A (ja)
CA (1) CA2240132C (ja)
MX (1) MX9804834A (ja)
NO (1) NO983043D0 (ja)
WO (1) WO1998019462A1 (ja)

Families Citing this family (188)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE69723325T2 (de) * 1996-03-15 2004-06-03 Matsushita Electric Industrial Co., Ltd., Kadoma Verfahren und Vorrichtung zur Codierung und Decodierung von Bildern
JP4034380B2 (ja) * 1996-10-31 2008-01-16 株式会社東芝 画像符号化/復号化方法及び装置
EP0940774A3 (en) * 1998-03-05 2000-07-05 Matsushita Electric Industrial Co., Ltd. Motion vector coding and decoding apparatus and method
JP3382173B2 (ja) * 1998-03-10 2003-03-04 株式会社ハイニックスセミコンダクター 飛越し走査方式の映像信号に対するvop境界矩形の設定方法及び装置
KR100282147B1 (ko) * 1998-11-13 2001-02-15 구자홍 압축 영상 복원 방법
US6563955B2 (en) * 1998-11-13 2003-05-13 Xerox Corporation Method and apparatus for analyzing image data to use multiple transforms for enhanced image data transmission
US6563953B2 (en) * 1998-11-30 2003-05-13 Microsoft Corporation Predictive image compression using a single variable length code for both the luminance and chrominance blocks for each macroblock
US6377713B1 (en) * 1999-01-27 2002-04-23 General Instrument Corporation Synchronous DRAM bandwidth optimization for display downsizing of an MPEG-2 image
JP2000333171A (ja) * 1999-05-12 2000-11-30 Neucore Technol Inc 画像処理装置
US7136525B1 (en) * 1999-09-20 2006-11-14 Microsoft Corporation System and method for background maintenance of an image sequence
US8131867B1 (en) 2000-06-01 2012-03-06 Qualcomm Incorporated Dynamic layer congestion control for multicast transport
US6486803B1 (en) 2000-09-22 2002-11-26 Digital Fountain, Inc. On demand encoding with a window
US6411223B1 (en) * 2000-10-18 2002-06-25 Digital Fountain, Inc. Generating high weight encoding symbols using a basis
JP3636983B2 (ja) 2000-10-23 2005-04-06 日本放送協会 符号化装置
JP2002152733A (ja) * 2000-11-08 2002-05-24 Fujitsu Ltd 動画像符号化方法、動画像復号化方法、動画像符号化装置、動画像復号化装置、動画像符号化プログラムを記録した記録媒体、動画像復号化プログラムを記録した記録媒体、および動画像符号化データを記録した記録媒体
US6901173B2 (en) * 2001-04-25 2005-05-31 Lockheed Martin Corporation Scene-based non-uniformity correction for detector arrays
US7103235B2 (en) * 2001-04-25 2006-09-05 Lockheed Martin Corporation Extended range image processing for electro-optical systems
US6973218B2 (en) * 2001-04-25 2005-12-06 Lockheed Martin Corporation Dynamic range compression
FR2830159A1 (fr) * 2001-09-24 2003-03-28 Thomson Licensing Sa Procede de codage selon la norme mpeg
WO2003043346A1 (en) * 2001-11-16 2003-05-22 Ntt Docomo, Inc. Image encoding method, image decoding method, image encoder, image decode, program, computer data signal, and image transmission system
ES2610430T3 (es) 2001-12-17 2017-04-27 Microsoft Technology Licensing, Llc Codificación por omisión de macrobloques
KR100556732B1 (ko) * 2001-12-29 2006-03-10 엘지전자 주식회사 동영상 확대영역 추적방법
US7978765B2 (en) * 2002-03-22 2011-07-12 Realnetworks, Inc. Context-adaptive macroblock type encoding/decoding methods and apparatuses
JP4090862B2 (ja) 2002-04-26 2008-05-28 松下電器産業株式会社 可変長符号化方法および可変長復号化方法
US7596279B2 (en) * 2002-04-26 2009-09-29 Ntt Docomo, Inc. Image encoding device, image decoding device, image encoding method, image decoding method, image encoding program, and image decoding program
JP3900017B2 (ja) * 2002-06-04 2007-04-04 富士ゼロックス株式会社 画像処理装置
US8406301B2 (en) 2002-07-15 2013-03-26 Thomson Licensing Adaptive weighting of reference pictures in video encoding
US7903742B2 (en) * 2002-07-15 2011-03-08 Thomson Licensing Adaptive weighting of reference pictures in video decoding
ZA200500417B (en) * 2002-07-15 2006-10-25 Thomson Licensing Sa Adaptive weighting of reference pictures in video decoding
US7376186B2 (en) * 2002-07-15 2008-05-20 Thomson Licensing Motion estimation with weighting prediction
US7266247B2 (en) * 2002-09-30 2007-09-04 Samsung Electronics Co., Ltd. Image coding method and apparatus using spatial predictive coding of chrominance and image decoding method and apparatus
US6931061B2 (en) * 2002-11-13 2005-08-16 Sony Corporation Method of real time MPEG-4 texture decoding for a multiprocessor environment
NO318167B1 (no) * 2002-11-27 2005-02-14 Tandberg Telecom As Vektorprediksjon
US9108107B2 (en) 2002-12-10 2015-08-18 Sony Computer Entertainment America Llc Hosting and broadcasting virtual events using streaming interactive video
US8964830B2 (en) 2002-12-10 2015-02-24 Ol2, Inc. System and method for multi-stream video compression using multiple encoding formats
US8366552B2 (en) 2002-12-10 2013-02-05 Ol2, Inc. System and method for multi-stream video compression
US10201760B2 (en) 2002-12-10 2019-02-12 Sony Interactive Entertainment America Llc System and method for compressing video based on detected intraframe motion
US8549574B2 (en) 2002-12-10 2013-10-01 Ol2, Inc. Method of combining linear content and interactive content compressed together as streaming interactive video
US9077991B2 (en) 2002-12-10 2015-07-07 Sony Computer Entertainment America Llc System and method for utilizing forward error correction with video compression
US8979655B2 (en) 2002-12-10 2015-03-17 Ol2, Inc. System and method for securely hosting applications
US20090118019A1 (en) 2002-12-10 2009-05-07 Onlive, Inc. System for streaming databases serving real-time applications used through streaming interactive video
US9192859B2 (en) 2002-12-10 2015-11-24 Sony Computer Entertainment America Llc System and method for compressing video based on latency measurements and other feedback
US9446305B2 (en) 2002-12-10 2016-09-20 Sony Interactive Entertainment America Llc System and method for improving the graphics performance of hosted applications
US9227139B2 (en) 2002-12-10 2016-01-05 Sony Computer Entertainment America Llc Virtualization system and method for hosting applications
US8711923B2 (en) 2002-12-10 2014-04-29 Ol2, Inc. System and method for selecting a video encoding format based on feedback data
US9314691B2 (en) 2002-12-10 2016-04-19 Sony Computer Entertainment America Llc System and method for compressing video frames or portions thereof based on feedback information from a client device
US9061207B2 (en) * 2002-12-10 2015-06-23 Sony Computer Entertainment America Llc Temporary decoder apparatus and method
US9138644B2 (en) * 2002-12-10 2015-09-22 Sony Computer Entertainment America Llc System and method for accelerated machine switching
US8526490B2 (en) 2002-12-10 2013-09-03 Ol2, Inc. System and method for video compression using feedback including data related to the successful receipt of video content
US20100166056A1 (en) * 2002-12-10 2010-07-01 Steve Perlman System and method for encoding video using a selected tile and tile rotation pattern
RU2225035C1 (ru) * 2003-04-21 2004-02-27 Общество с ограниченной ответственностью "Р.Т.С.-Сервис" Способ кодирования координат перемещающегося на экране вычислительного устройства видеоизображения, устройство для декодирования визуального объекта, закодированного этим способом, и система, предназначенная для визуализации активного видео с помощью этого устройства
JP2005057738A (ja) * 2003-07-18 2005-03-03 Canon Inc 信号処理装置、信号処理方法及びプログラム
FR2858741A1 (fr) * 2003-08-07 2005-02-11 Thomson Licensing Sa Dispositif et procede de compression d'images numeriques
US7724827B2 (en) * 2003-09-07 2010-05-25 Microsoft Corporation Multi-layer run level encoding and decoding
US7606308B2 (en) * 2003-09-07 2009-10-20 Microsoft Corporation Signaling macroblock mode information for macroblocks of interlaced forward-predicted fields
US7860326B2 (en) * 2003-09-22 2010-12-28 Kddi Corporation Adaptable shape image encoding apparatus and decoding apparatus
DE602004017689D1 (de) * 2003-11-21 2008-12-24 Samsung Electronics Co Ltd Vorrichtung und Methode zur Erzeugung von kodierten Blockanordnungen für ein Alpha-Kanalbild sowie Alpha-Kanalkodierungs und -dekodierungsvorrichtung und -methode.
US20050111545A1 (en) 2003-11-25 2005-05-26 Ram Prabhakar Dynamic packet size control for MPEG-4 data partition mode
EP1538844A3 (en) * 2003-11-26 2006-05-31 Samsung Electronics Co., Ltd. Color image residue transformation and encoding method
US6864812B1 (en) * 2004-02-05 2005-03-08 Broadcom Corporation Hardware efficient implementation of finite impulse response filters with limited range input signals
JP2005223631A (ja) * 2004-02-05 2005-08-18 Sony Corp データ処理装置およびその方法と符号化装置および復号装置
US20050210008A1 (en) * 2004-03-18 2005-09-22 Bao Tran Systems and methods for analyzing documents over a network
US7590059B2 (en) * 2004-05-21 2009-09-15 Broadcom Corp. Multistandard video decoder
BRPI0515943B1 (pt) * 2004-09-29 2018-10-16 Thomson Res Funding Corporation método e aparelho para codificação e decodificação de vídeo de atualização de resolução reduzida
KR100647294B1 (ko) * 2004-11-09 2006-11-23 삼성전자주식회사 화상 데이터 부호화 및 복호화 방법 및 장치
US7650031B2 (en) * 2004-11-23 2010-01-19 Microsoft Corporation Method and system for detecting black frames in a sequence of frames
WO2006097144A1 (en) * 2005-03-15 2006-09-21 Nero Ag Apparatus and method for producing a picture data stream and for processing a picture data stream
EP1703734A1 (en) * 2005-03-15 2006-09-20 Nero AG Apparatus and method for producing and processing a picture data stream
TWI280805B (en) * 2005-07-20 2007-05-01 Novatek Microelectronics Corp Method and apparatus for cost calculation in decimal motion estimation
EP1911278A2 (en) * 2005-08-04 2008-04-16 Nds Limited Advanced digital tv system
US9077960B2 (en) * 2005-08-12 2015-07-07 Microsoft Corporation Non-zero coefficient block pattern coding
US8107748B2 (en) * 2005-09-16 2012-01-31 Sony Corporation Adaptive motion search range
US8005308B2 (en) * 2005-09-16 2011-08-23 Sony Corporation Adaptive motion estimation for temporal prediction filter over irregular motion vector samples
US7885335B2 (en) * 2005-09-16 2011-02-08 Sont Corporation Variable shape motion estimation in video sequence
US7894522B2 (en) * 2005-09-16 2011-02-22 Sony Corporation Classified filtering for temporal prediction
US8059719B2 (en) * 2005-09-16 2011-11-15 Sony Corporation Adaptive area of influence filter
US7957466B2 (en) * 2005-09-16 2011-06-07 Sony Corporation Adaptive area of influence filter for moving object boundaries
US7596243B2 (en) * 2005-09-16 2009-09-29 Sony Corporation Extracting a moving object boundary
US8165205B2 (en) * 2005-09-16 2012-04-24 Sony Corporation Natural shaped regions for motion compensation
US7894527B2 (en) * 2005-09-16 2011-02-22 Sony Corporation Multi-stage linked process for adaptive motion vector sampling in video compression
US20070064805A1 (en) * 2005-09-16 2007-03-22 Sony Corporation Motion vector selection
US7620108B2 (en) * 2005-09-16 2009-11-17 Sony Corporation Integrated spatial-temporal prediction
KR101455578B1 (ko) 2005-09-26 2014-10-29 미쓰비시덴키 가부시키가이샤 동화상 부호화 장치 및 동화상 복호 장치
KR20070037248A (ko) * 2005-09-30 2007-04-04 삼성전자주식회사 영상 부호화장치 및 방법, 영상 복호화장치 및 방법과 이를채용한 디스플레이 구동회로 및 방법
TWI256776B (en) * 2005-10-25 2006-06-11 Novatek Microelectronics Corp Method and apparatus for calculating cost functions and the interpolation method thereof
JP4555212B2 (ja) * 2005-11-10 2010-09-29 株式会社沖データ 透かし情報埋め込み装置,透かし情報埋め込み方法,およびコンピュータプログラム
KR100772873B1 (ko) * 2006-01-12 2007-11-02 삼성전자주식회사 스무딩 예측을 이용한 다계층 기반의 비디오 인코딩 방법,디코딩 방법, 비디오 인코더 및 비디오 디코더
US8705630B2 (en) * 2006-02-10 2014-04-22 Nvidia Corporation Adapting one type of encoder to another type of encoder
KR101311403B1 (ko) * 2006-07-04 2013-09-25 삼성전자주식회사 영상의 부호화 방법 및 장치, 복호화 방법 및 장치
US20080065980A1 (en) * 2006-09-08 2008-03-13 Opera Software Asa Modifying a markup language document which includes a clickable image
EP2067120B1 (en) * 2006-09-26 2016-07-27 AMBX UK Limited Creation and handling of a bitstream comprising video frames and auxiliary data
US20080212895A1 (en) * 2007-01-09 2008-09-04 Lockheed Martin Corporation Image data processing techniques for highly undersampled images
US8594186B1 (en) * 2007-02-27 2013-11-26 Xilinx, Inc. Digital video coding using quantized DC block values
US8594176B2 (en) * 2007-03-06 2013-11-26 Microsoft Corporation Streaming media codec with transform coefficient bounding
TW200837663A (en) * 2007-03-13 2008-09-16 Univ Nat Taiwan Constant picture quality bit rate control system for a multiple-video encoder in single video signal source and the method
TWI339531B (en) * 2007-05-07 2011-03-21 Vivotek Inc A module and system for generating different resolution video streams immediately
JP5092558B2 (ja) * 2007-06-08 2012-12-05 株式会社日立製作所 画像符号化方法、画像符号化装置、画像復号化方法及び画像復号化装置
US20080316221A1 (en) * 2007-06-19 2008-12-25 Aas Eric F Down-sampled image display
JP4325708B2 (ja) * 2007-07-05 2009-09-02 ソニー株式会社 データ処理装置、データ処理方法およびデータ処理プログラム、符号化装置、符号化方法および符号化プログラム、ならびに、復号装置、復号方法および復号プログラム
US7895347B2 (en) * 2007-07-27 2011-02-22 Red Hat, Inc. Compact encoding of arbitrary length binary objects
WO2009032255A2 (en) * 2007-09-04 2009-03-12 The Regents Of The University Of California Hierarchical motion vector processing method, software and devices
US9168457B2 (en) 2010-09-14 2015-10-27 Sony Computer Entertainment America Llc System and method for retaining system state
ES2812473T3 (es) * 2008-03-19 2021-03-17 Nokia Technologies Oy Vector de movimiento combinado y predicción de índice de referencia para la codificación de vídeo
US8208543B2 (en) * 2008-05-19 2012-06-26 Microsoft Corporation Quantization and differential coding of alpha image data
WO2009148076A1 (ja) 2008-06-05 2009-12-10 日本電信電話株式会社 映像符号量制御方法、映像符号量制御装置、映像符号量制御プログラムおよびそのプログラムを記録したコンピュータ読み取り可能な記録媒体
KR101364195B1 (ko) 2008-06-26 2014-02-21 에스케이텔레콤 주식회사 움직임벡터 부호화/복호화 방법 및 그 장치
JP5029543B2 (ja) * 2008-09-09 2012-09-19 富士通株式会社 符号化装置および符号化プログラム
CN105721877A (zh) 2008-09-23 2016-06-29 杜比实验室特许公司 对交织的图像数据进行解码的方法和包括编解码器的装置
KR101543298B1 (ko) * 2008-10-13 2015-08-10 에스케이 텔레콤주식회사 동영상 부호화/복호화 장치 및 그를 위한 가변 단위의 적응적 중첩 블록 움직임 보상 장치 및 방법
CN105357509B (zh) * 2009-01-29 2017-09-15 杜比实验室特许公司 视频编码方法、视频信号解码方法及视频装置
WO2010090629A1 (en) * 2009-02-05 2010-08-12 Thomson Licensing Methods and apparatus for adaptive mode video encoding and decoding
US9729899B2 (en) 2009-04-20 2017-08-08 Dolby Laboratories Licensing Corporation Directed interpolation and data post-processing
US9165605B1 (en) * 2009-09-11 2015-10-20 Lindsay Friedman System and method for personal floating video
US20110137969A1 (en) * 2009-12-09 2011-06-09 Mangesh Sadafale Apparatus and circuits for shared flow graph based discrete cosine transform
KR101785666B1 (ko) * 2010-01-12 2017-10-16 엘지전자 주식회사 비디오 신호의 처리 방법 및 장치
MX2012009474A (es) * 2010-02-24 2012-10-09 Sharp Kk Dispositivo de codificacion de imagen y dispositivo de decodificacion de imagen.
US20120320966A1 (en) * 2010-03-09 2012-12-20 Telegent Systems Inc. c/o M & C Corporate Services Limited Adaptive video decoding circuitry and techniques
US9601692B1 (en) 2010-07-13 2017-03-21 Crossbar, Inc. Hetero-switching layer in a RRAM device and method
US8946046B1 (en) 2012-05-02 2015-02-03 Crossbar, Inc. Guided path for forming a conductive filament in RRAM
US9570678B1 (en) 2010-06-08 2017-02-14 Crossbar, Inc. Resistive RAM with preferental filament formation region and methods
US9012307B2 (en) 2010-07-13 2015-04-21 Crossbar, Inc. Two terminal resistive switching device structure and method of fabricating
CN103081093B (zh) 2010-06-11 2015-06-03 科洛斯巴股份有限公司 存储器件的柱结构以及方法
JP5012967B2 (ja) * 2010-07-05 2012-08-29 カシオ計算機株式会社 画像処理装置及び方法、並びにプログラム
US8374018B2 (en) 2010-07-09 2013-02-12 Crossbar, Inc. Resistive memory using SiGe material
US8569172B1 (en) 2012-08-14 2013-10-29 Crossbar, Inc. Noble metal/non-noble metal electrode for RRAM applications
US8168506B2 (en) 2010-07-13 2012-05-01 Crossbar, Inc. On/off ratio for non-volatile memory device and method
US8947908B2 (en) 2010-11-04 2015-02-03 Crossbar, Inc. Hetero-switching layer in a RRAM device and method
US8884261B2 (en) 2010-08-23 2014-11-11 Crossbar, Inc. Device switching using layered device structure
JP5706771B2 (ja) * 2010-07-30 2015-04-22 キヤノン株式会社 動画像の予測符号化装置及びその制御方法、コンピュータプログラム
US9401475B1 (en) 2010-08-23 2016-07-26 Crossbar, Inc. Method for silver deposition for a non-volatile memory device
US8889521B1 (en) 2012-09-14 2014-11-18 Crossbar, Inc. Method for silver deposition for a non-volatile memory device
US8492195B2 (en) 2010-08-23 2013-07-23 Crossbar, Inc. Method for forming stackable non-volatile resistive switching memory devices
US8558212B2 (en) 2010-09-29 2013-10-15 Crossbar, Inc. Conductive path in switching material in a resistive random access memory device and control
USRE46335E1 (en) 2010-11-04 2017-03-07 Crossbar, Inc. Switching device having a non-linear element
US8502185B2 (en) 2011-05-31 2013-08-06 Crossbar, Inc. Switching device having a non-linear element
WO2012077349A1 (ja) * 2010-12-09 2012-06-14 パナソニック株式会社 画像符号化方法および画像復号化方法
US8930174B2 (en) 2010-12-28 2015-01-06 Crossbar, Inc. Modeling technique for resistive random access memory (RRAM) cells
US9153623B1 (en) 2010-12-31 2015-10-06 Crossbar, Inc. Thin film transistor steering element for a non-volatile memory device
US8815696B1 (en) 2010-12-31 2014-08-26 Crossbar, Inc. Disturb-resistant non-volatile memory device using via-fill and etchback technique
US9620206B2 (en) 2011-05-31 2017-04-11 Crossbar, Inc. Memory array architecture with two-terminal memory cells
US8619459B1 (en) 2011-06-23 2013-12-31 Crossbar, Inc. High operating speed resistive random access memory
US9627443B2 (en) 2011-06-30 2017-04-18 Crossbar, Inc. Three-dimensional oblique two-terminal memory with enhanced electric field
HUE052633T2 (hu) * 2011-06-30 2021-05-28 Jvc Kenwood Corp Képdekódoló készülék, képdekódoló eljárás és képdekódoló program
US8946669B1 (en) 2012-04-05 2015-02-03 Crossbar, Inc. Resistive memory device and fabrication methods
US9564587B1 (en) 2011-06-30 2017-02-07 Crossbar, Inc. Three-dimensional two-terminal memory with enhanced electric field and segmented interconnects
US9166163B2 (en) 2011-06-30 2015-10-20 Crossbar, Inc. Sub-oxide interface layer for two-terminal memory
US9252191B2 (en) 2011-07-22 2016-02-02 Crossbar, Inc. Seed layer for a p+ silicon germanium material for a non-volatile memory device and method
US10056907B1 (en) 2011-07-29 2018-08-21 Crossbar, Inc. Field programmable gate array utilizing two-terminal non-volatile memory
US9729155B2 (en) 2011-07-29 2017-08-08 Crossbar, Inc. Field programmable gate array utilizing two-terminal non-volatile memory
US8674724B2 (en) 2011-07-29 2014-03-18 Crossbar, Inc. Field programmable gate array utilizing two-terminal non-volatile memory
JP2013158851A (ja) * 2012-02-01 2013-08-19 Seiko Epson Corp ヤコビ行列情報生成装置、ヤコビ行列情報生成方法、ヤコビ行列情報生成プログラム、ロボット制御装置、ロボット制御方法、ロボット制御プログラムおよびロボットシステム
US9087576B1 (en) 2012-03-29 2015-07-21 Crossbar, Inc. Low temperature fabrication method for a three-dimensional memory device and structure
US9685608B2 (en) 2012-04-13 2017-06-20 Crossbar, Inc. Reduced diffusion in metal electrode for two-terminal memory
US8658476B1 (en) 2012-04-20 2014-02-25 Crossbar, Inc. Low temperature P+ polycrystalline silicon material for non-volatile memory device
US8796658B1 (en) 2012-05-07 2014-08-05 Crossbar, Inc. Filamentary based non-volatile resistive memory device and method
SG11201407417VA (en) 2012-05-14 2014-12-30 Luca Rossato Encoding and reconstruction of residual data based on support information
US9583701B1 (en) 2012-08-14 2017-02-28 Crossbar, Inc. Methods for fabricating resistive memory device switching material using ion implantation
US10096653B2 (en) 2012-08-14 2018-10-09 Crossbar, Inc. Monolithically integrated resistive memory using integrated-circuit foundry compatible processes
US8946673B1 (en) 2012-08-24 2015-02-03 Crossbar, Inc. Resistive switching device structure with improved data retention for non-volatile memory device and method
US9312483B2 (en) 2012-09-24 2016-04-12 Crossbar, Inc. Electrode structure for a non-volatile memory device and method
US9576616B2 (en) 2012-10-10 2017-02-21 Crossbar, Inc. Non-volatile memory with overwrite capability and low write amplification
US8982647B2 (en) 2012-11-14 2015-03-17 Crossbar, Inc. Resistive random access memory equalization and sensing
US8942491B2 (en) * 2012-11-29 2015-01-27 Konica Minolta Laboratory U.S.A., Inc. Topology-preserving downsampling of binary images
US9412790B1 (en) 2012-12-04 2016-08-09 Crossbar, Inc. Scalable RRAM device architecture for a non-volatile memory device and method
TWI571828B (zh) * 2013-01-02 2017-02-21 奇高電子股份有限公司 光學導航方法以及相關裝置
US9406379B2 (en) 2013-01-03 2016-08-02 Crossbar, Inc. Resistive random access memory with non-linear current-voltage relationship
US9112145B1 (en) 2013-01-31 2015-08-18 Crossbar, Inc. Rectified switching of two-terminal memory via real time filament formation
US9324942B1 (en) 2013-01-31 2016-04-26 Crossbar, Inc. Resistive memory cell with solid state diode
GB2512658B (en) * 2013-04-05 2020-04-01 British Broadcasting Corp Transmitting and receiving a composite image
US9544613B2 (en) * 2013-04-24 2017-01-10 Sony Corporation Local detection model (LDM) for recursive motion estimation
US20160165246A1 (en) * 2013-07-17 2016-06-09 Sony Corporation Image processing device and method
GB2538856B (en) * 2014-02-06 2018-04-25 Imagination Tech Ltd Opacity testing for processing primitives in a 3D graphics processing system
US10290801B2 (en) 2014-02-07 2019-05-14 Crossbar, Inc. Scalable silicon based resistive memory device
KR101551915B1 (ko) * 2015-03-16 2015-09-08 스타십벤딩머신 주식회사 영상압축방법 및 영상압축장치
KR102119300B1 (ko) 2017-09-15 2020-06-04 서울과학기술대학교 산학협력단 360도 영상 부호화 장치 및 방법, 이를 수행하기 위한 기록 매체
KR102568633B1 (ko) * 2018-01-26 2023-08-21 삼성전자주식회사 이미지 처리 장치
JP7180079B2 (ja) * 2018-02-28 2022-11-30 セイコーエプソン株式会社 回路装置及び電子機器
WO2019185983A1 (en) * 2018-03-28 2019-10-03 Nokia Technologies Oy A method, an apparatus and a computer program product for encoding and decoding digital volumetric video
WO2020080698A1 (ko) 2018-10-19 2020-04-23 삼성전자 주식회사 영상의 주관적 품질을 평가하는 방법 및 장치
WO2020080765A1 (en) 2018-10-19 2020-04-23 Samsung Electronics Co., Ltd. Apparatuses and methods for performing artificial intelligence encoding and artificial intelligence decoding on image
KR102525578B1 (ko) 2018-10-19 2023-04-26 삼성전자주식회사 부호화 방법 및 그 장치, 복호화 방법 및 그 장치
WO2020080665A1 (en) 2018-10-19 2020-04-23 Samsung Electronics Co., Ltd. Methods and apparatuses for performing artificial intelligence encoding and artificial intelligence decoding on image
WO2020080827A1 (en) 2018-10-19 2020-04-23 Samsung Electronics Co., Ltd. Ai encoding apparatus and operation method of the same, and ai decoding apparatus and operation method of the same
WO2020080873A1 (en) 2018-10-19 2020-04-23 Samsung Electronics Co., Ltd. Method and apparatus for streaming data
US11317093B2 (en) 2019-09-24 2022-04-26 Tencent America LLC Method for reference picture resampling with offset in video bitstream
KR102436512B1 (ko) 2019-10-29 2022-08-25 삼성전자주식회사 부호화 방법 및 그 장치, 복호화 방법 및 그 장치
KR20210056179A (ko) 2019-11-08 2021-05-18 삼성전자주식회사 Ai 부호화 장치 및 그 동작방법, 및 ai 복호화 장치 및 그 동작방법
KR102287942B1 (ko) 2020-02-24 2021-08-09 삼성전자주식회사 전처리를 이용한 영상의 ai 부호화 및 ai 복호화 방법, 및 장치

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0389792A (ja) * 1989-09-01 1991-04-15 Hitachi Ltd 画像符号化装置
JPH08116542A (ja) * 1994-10-19 1996-05-07 Matsushita Electric Ind Co Ltd 画像符号化装置、画像復号化装置および動きベクトル検出装置
JPH08214318A (ja) * 1994-10-13 1996-08-20 At & T Corp ビデオ画像のシーケンスを符号化するための領域ベースアプローチに対する方法及び装置

Family Cites Families (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS4982219A (ja) 1972-12-11 1974-08-08
JPS6132664A (ja) 1984-07-24 1986-02-15 Fuji Xerox Co Ltd 画像デ−タの圧縮方法
JPH01220989A (ja) 1989-01-27 1989-09-04 Nec Corp 多値画像信号の適応予測復号化装置
JPH03117991A (ja) * 1989-09-29 1991-05-20 Victor Co Of Japan Ltd 動きベクトル符号化装置及び復号化装置
JP2897082B2 (ja) 1991-03-27 1999-05-31 富士通株式会社 動き補償可変長符号化方式
WO1993018618A1 (en) * 1992-03-03 1993-09-16 Kabushiki Kaisha Toshiba Time-varying image encoder
US5714950A (en) * 1992-07-23 1998-02-03 Samsung Electronics Co., Ltd. System for variable-length-coding and variable-length-decoding digitaldata
JP3545000B2 (ja) * 1992-11-02 2004-07-21 ソニー株式会社 画像信号符号化装置、画像信号復号化装置
JPH07152915A (ja) 1993-11-29 1995-06-16 Toshiba Corp 画像符号化装置及び復合化装置
DE69533870T2 (de) * 1994-10-19 2005-05-25 Matsushita Electric Industrial Co., Ltd., Kadoma Vorrichtung zur Bilddekodierung
EP0709809B1 (en) * 1994-10-28 2002-01-23 Oki Electric Industry Company, Limited Image encoding and decoding method and apparatus using edge synthesis and inverse wavelet transform
JPH08294119A (ja) * 1995-04-21 1996-11-05 Toshiba Corp 画像符号化/復号化装置
KR100235345B1 (ko) 1994-12-29 1999-12-15 전주범 분할영역에서의 움직임 추정방법 및 장치
JP3169783B2 (ja) 1995-02-15 2001-05-28 日本電気株式会社 動画像の符号化・復号システム
KR100209409B1 (ko) * 1995-03-20 1999-07-15 전주범 물체 왜곡을 이용한 물체 기반 부호화 방법 및 장치
JPH08265755A (ja) 1995-03-22 1996-10-11 Canon Inc 画像処理装置および画像処理方法
US5883678A (en) * 1995-09-29 1999-03-16 Kabushiki Kaisha Toshiba Video coding and video decoding apparatus for reducing an alpha-map signal at a controlled reduction ratio
US5764814A (en) * 1996-03-22 1998-06-09 Microsoft Corporation Representation and encoding of general arbitrary shapes
KR100209412B1 (ko) * 1996-05-10 1999-07-15 전주범 비디오 신호의 유호 색차 성분 부호화 방법
KR100209413B1 (ko) * 1996-05-23 1999-07-15 전주범 블럭-기반 비디오 신호 부호화 시스템에 이용하기 위한그리드 결정방법
US5881175A (en) * 1996-06-07 1999-03-09 Daewoo Electronics Co., Ltd. Method and apparatus for encoding an image signal by using the contour signal thereof
JP3210862B2 (ja) * 1996-06-27 2001-09-25 シャープ株式会社 画像符号化装置及び画像復号装置
KR100209419B1 (ko) * 1996-07-09 1999-07-15 전주범 영상신호로 표현된 객체의 윤곽선 부호화 방법
JP4034380B2 (ja) * 1996-10-31 2008-01-16 株式会社東芝 画像符号化/復号化方法及び装置
WO1998031151A1 (en) * 1997-01-10 1998-07-16 Matsushita Electric Industrial Co., Ltd. Image processing method, image processing device, and data recording medium
US6014466A (en) * 1997-07-10 2000-01-11 Hughes Electronics Corporation Object-based video coding of arbitrarily shaped objects using lapped orthogonal transforms (LOTs) defined on rectangular and L-shaped regions
KR100295798B1 (ko) * 1997-07-11 2001-08-07 전주범 스케일러빌리티를구현한이진현상신호부호화장치

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0389792A (ja) * 1989-09-01 1991-04-15 Hitachi Ltd 画像符号化装置
JPH08214318A (ja) * 1994-10-13 1996-08-20 At & T Corp ビデオ画像のシーケンスを符号化するための領域ベースアプローチに対する方法及び装置
JPH08116542A (ja) * 1994-10-19 1996-05-07 Matsushita Electric Ind Co Ltd 画像符号化装置、画像復号化装置および動きベクトル検出装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP0873017A4 *

Also Published As

Publication number Publication date
US20050053154A1 (en) 2005-03-10
US20070147514A1 (en) 2007-06-28
EP0873017A1 (en) 1998-10-21
CA2240132C (en) 2002-09-03
CA2240132A1 (en) 1998-05-07
JP4034380B2 (ja) 2008-01-16
CN1139256C (zh) 2004-02-18
US6292514B1 (en) 2001-09-18
AU713780B2 (en) 1999-12-09
US6198768B1 (en) 2001-03-06
MX9804834A (es) 1998-12-31
US7167521B2 (en) 2007-01-23
NO983043L (no) 1998-06-30
BR9706905A (pt) 1999-07-20
CN1207229A (zh) 1999-02-03
KR19990076787A (ko) 1999-10-15
US20040091049A1 (en) 2004-05-13
US6754269B1 (en) 2004-06-22
CN100382603C (zh) 2008-04-16
CN1516476A (zh) 2004-07-28
US20050084016A1 (en) 2005-04-21
US7215709B2 (en) 2007-05-08
AU4726697A (en) 1998-05-22
KR100286443B1 (ko) 2001-04-16
US6122318A (en) 2000-09-19
US7308031B2 (en) 2007-12-11
JPH1155667A (ja) 1999-02-26
EP0873017A4 (en) 2000-05-24
NO983043D0 (no) 1998-06-30

Similar Documents

Publication Publication Date Title
WO1998019462A1 (en) Image encoder and image decoder
KR100289587B1 (ko) 화상 신호 부호화 방법 및 화상 신호 부호화 장치, 화상 신호 복호화 방법 및 화상 신호 복호화장치,및 화상 신호 기록매체
US7298911B2 (en) Video coding and video decoding apparatus
US6571017B1 (en) Method and apparatus for digital image coding and decoding, and data recording media
JPH09182084A (ja) 動画像符号化装置および動画像復号化装置
JPH06209468A (ja) 画像信号符号化方法および画像信号符号化装置、並びに画像信号復号化方法および画像信号復号化装置
US6259738B1 (en) Video encoding apparatus and video decoding apparatus
EP2076050B1 (en) Motion estimation apparatus and method for moving picture coding
JP2008228336A (ja) 画像符号化方法及び装置
JPH104549A (ja) 画像符号化装置および画像復号化装置
JP2004343788A (ja) 画像符号化装置および画像復号化装置および画像符号化データを記録した記録媒体
JP2002051347A (ja) 動画像符号化装置
JP2004350307A (ja) 画像符号化装置および画像復号化装置
JP4193252B2 (ja) 信号処理装置及び方法、信号復号装置、並びに信号符号化装置
JP2004320807A (ja) 画像符号化装置および画像復号化装置
JP2004320808A (ja) 画像符号化装置
JP2008011564A (ja) 画像符号化装置および画像復号化装置
JPH07336681A (ja) 画像変換装置
Turaga et al. ITU-T Video Coding Standards
JPH03295378A (ja) 画像符号化装置
JPH1141600A (ja) 画像階層符号化・復号方法および装置並びに記録媒体
Hu Image and video scalability in the compressed domain.
JPH10191339A (ja) 画像信号の補填方法及び画像信号復号化装置

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 97191547.4

Country of ref document: CN

AK Designated states

Kind code of ref document: A1

Designated state(s): AU BR CA CN KR MX NO SG US VN

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): CH DE ES FR GB IT NL SE

ENP Entry into the national phase

Country of ref document: CA

Ref document number: 2240132

Country of ref document: CA

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: PA/a/1998/004834

Country of ref document: MX

WWE Wipo information: entry into national phase

Ref document number: 1997909712

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 09091362

Country of ref document: US

WWE Wipo information: entry into national phase

Ref document number: 1019980704912

Country of ref document: KR

WWE Wipo information: entry into national phase

Ref document number: 1199800651

Country of ref document: VN

121 Ep: the epo has been informed by wipo that ep was designated in this application
WWP Wipo information: published in national office

Ref document number: 1997909712

Country of ref document: EP

WWP Wipo information: published in national office

Ref document number: 1019980704912

Country of ref document: KR

WWG Wipo information: grant in national office

Ref document number: 1019980704912

Country of ref document: KR