CROSS-REFERENCES TO RELATED PATENT APPLICATIONS
FIELD OF THE INVENTION
This is a Continuation of application Ser. No. 10/612,013, filed Jul. 3, 2003; which is a Continuation of application Ser. No. 09/703,649, filed Nov. 2, 2000; which is a Continuation of application Ser. No. 08/024,305, filed Mar. 1, 1993; the disclosures of which are incorporated herein by reference.
- BACKGROUND OF THE INVENTION
The present invention relates to a signal compressing system. A system according to the present invention is particularly suited for compressing image signals. The present disclosure is based on the disclosure in Korean Patent Application No. 92-3398 filed Feb. 29, 1992, which disclosure is incorporated herein by reference.
Image signals may be compressed by motion-compensated interframe discrete cosine transform (DCT) coding such as is defined by a MPEG (Moving Picture Expert Group) international standard. This form of signal compression has attracted much attention in the field of high definition television (HDTV).
FIG. 1 is a block diagram of such a conventional motion-compensated interframe DCT coder. In the shown coder, an image signal is divided into a plurality of sub-blocks. The sub-blocks are all of the same size, for example 8×8, 16×16, . . . . A motion estimator 40 produces a motion vector, defined by the difference between the current image signal and a one-frame delayed image signal, output by a frame memory 30. The motion vector is supplied to a motion compensator 50 which compensates the delayed image signal from the frame memory 30 on the basis of the motion vector. A first adder 8 a serves to produce the difference between the present frame and the delayed, motion compensated frame. A discrete cosine transform portion 10 processes the difference signal, output by the first adder 8 a, for a sub-block. The motion estimator 40 determines the motion vector by using a block matching algorithm.
The discrete cosine transformed signal is quantized by a quantizer 20. The image signal is scanned in a zig-zag manner to produce a runlength coded version thereof. The runlength coded signal comprises a plurality of strings which include a series of “0”s, representing the run length, and an amplitude value of any value except “0”.
The runlength coded signal is dequantized by a dequantizer 21, inversely zig-zag scanned and inversely discrete cosine transformed by an inverse discrete cosine transforming portion 11. The transformed image signal is added to the motion-compensated estimate error signal by a second adder 8 b. As a result the image signal is decoded into a signal corresponding to the original image signal.
Refresh switches RSW1, RSW2 are arranged between the adders 8 a, 8 b and the motion compensator 40 so as to provide the original image signal free from externally induced errors.
The runlength coded signal is also supplied to a variable length coder 60 which applies a variable length coding to the runlength coded image signal. The variable length coded signal is then output through a FIFO transfer buffer 70 as a coded image signal.
In motion-compensated adaptive DCT coding, the interframe signal can be easily estimated or coded by way of motion compensation, thereby obtaining a high coding efficiency, since the image signal has a relatively high correlation along the time axis. That is, according to the afore-mentioned method, the coding efficiency is high because most of the energy of a discrete cosine transformed signal is compressed at the lower end of its spectrum, resulting in long runs of “0”s in the runlength coded signal.
However, the scanning regime of the aforementioned method does not take account of differences in the spectrum of the motion-compensated interframe DCT signal with time.
- SUMMARY OF THE INVENTION
A method is known wherein one of a plurality of reference modes is previously selected on the basis of the difference between the present block and that of a previous frame and the image signal is scanned by way of a scanning pattern under the selected mode and suitably quantized. With such a method, however, three modes are employed to compute the energies of the intermediate and high frequency components of the image signal in accordance with the interframe or the intraframe modes in order to determine the appropriate mode. This mode determining procedure is undesirably complicated.
According to the present invention, there is provided a signal compressing system, comprising coding means for scanning an input signal according to a plurality of different scanning patterns to provided coded versions thereof and selection means for selecting a said scanning pattern which produces efficient coding according to a predetermined criterion and outputting a scanning pattern signal identifying the selected scanning pattern.
Preferably, the input signal is an inherently two-dimensional signal, for example, an image signal.
Preferably, the coding means codes the input signal according to a runlength coding regime.
Preferably, the system includes a variable length coder to variably length code the coded signal, produced by scanning according to the selected scanning pattern.
BRIEF DESCRIPTION OF THE DRAWINGS
Preferably, the system includes discrete cosine transformer means to produce said input signal. The transformer means may be a motion-compensated interframe adaptive discrete cosine transformer.
An embodiment of the present invention will now be described, by way of example, with reference to FIGS. 2 and 3 of the accompanying drawings, in which:
FIG. 1 is a block diagram of a conventional adaptive interframe DCT coding system employing a motion compensating technique;
FIG. 2 is a block diagram of a coding system embodying the present invention;
FIGS. 3A-3H show various possible scanning patterns according to the present invention; and
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
FIG. 4 is a block diagram of a decoding system according to the present invention.
Referring to FIG. 2, an input signal is divided into equal-sized sub-blocks, for example, 8×8, 16×16, . . . . A motion estimator 40 determines a motion vector by comparing the current frame and a one frame delayed signal from a frame memory 30.
The motion vector is supplied to a motion compensator 60 which, in turn, compensates the delayed frame signal for movement. A first adder 8 a produces a difference signal representing the difference between the present frame and the delayed, motion-compensated frame. A DCT coder 10 DCT-codes the difference signal. The DCT coded image signal is quantized by a quantizer 20 and then dequantized by a dequantizer 21. The dequantized signal is supplied to a second adder 8 b, via IDCT 11, which adds it to the output of the motion compensator 11. This produces a signal corresponding to the original image signal.
The output of the motion compensator 50 is applied to the adders 8 a, 8 b by refresh switches RSW2 and RSW1, respectively.
The quantized image signal is also supplied to a multi-scanner 80 which scans it according to a plurality of predetermined patterns.
A scanner pattern selector 90 selects the scanning pattern which produces the minimum number of bits to represent the current sub-block. The scanning pattern selector also produces selection data which identifies the selected scanning pattern.
The image signal output by the scanning pattern selector 90 is variable length coded by a variable length coder 60. The variable length coder 60 compresses the image signal output by the scanning pattern selector 90. The variable length coder 60 operates such that a large proportion of the data samples are each represented by a small number of bits while a small proportion of the data samples are each represented by a large number of bits.
When a discrete cosine transformed image signal is quantized and runlength coded, the number of “0”s is increased over all, while the number of “0”s decreases as the magnitude of the signal increases. Accordingly, data compression is achieved because “0” can be represented by only a few bits and “255” can be represented by a relatively large number of bits.
Both the variable length coded signal and the selection data are supplied to a multiplexer MUX1 which multiplexes the variable length coded signal and the selection data, and optionally additional information such as teletext.
Since the variable length coded signal has data words of different lengths, a transfer buffer 70 is employed to temporarily store the multiplexed signal and output it at a constant rate.
FIG. 4 shows a decoding system at a remote station that receives and extracts the encoded data. In FIG. 4, demultiplexer 100 receives coded data and, in an operation inverse to that performed at the coding system, extracts the variable length encoded data, the scanning pattern information and the additional information that had been multiplexed together at the coding system. Variable length decoder 110 variable length decodes the variable length encoded data, and scanner 120 receives the variable length decoded data and reconstructs the original sub-block using a scanning pattern indicated by the extracted scanning pattern selection signal. The scanner would necessarily have to select one from a plurality pattern that was available for encoding. Using components having the same margin as dequantizers 21 and IDCT 11 in the encoder system, dequantizer 120 dequantizes the signal output from the scanner 120, and inverse discrete cosine transformer 140 performs an inverse discrete cosine transform function on the output of dequantizer 130, to output decoded data.
The original image signal is reconstructed at a remote station by performing the appropriate inverse scanning of the runlength coded signal in accordance with the multiplexed scanning pattern selection data.
FIGS. 3A to 3H show possible scanning patterns employed by the multi-scanner 80. Additional scanning patterns will be apparent to those skilled in the art. However, if the number of patterns becomes too large, the coding efficiency is degraded as the selection data word becomes longer.
As described above, according to the present invention, the quantized image signal is scanned according to various scanning patterns, and then the most efficient pattern is selected.
A suitable measure of efficiency is the number of bits required to runlength code the image signal.