CA2520652A1 - Combined runlength coding and variable length coding for video compression - Google Patents

Abstract

A method and an apparatus to encode a series of quantized coefficients of a transform of a block of image data. The transform is such that 0 is the most likely coefficient amplitude and 1 is the next most likely coefficient amplitude. The method includes forming modes for each encountered non-zero amplitude and forming events for the modes in the mode-~modified series formed by setting to zero all non-zero amplitudes other than the amplitude defining the mode. An event for modes for amplitudes greater than 1 is a run of none or more zero-valued coefficients followed a non-zero amplitude in the mode-modified series. In one embodiment, a first mode is defined for amplitude 1 wherein an event is a run of none or more zero-valued coefficients followed the amplitude 1 coefficient in the mode-modified series. In another embodiment, a first mode is defined for amplitude 1 wherein an event is a run in the mode-modified series of none or more zero-valued coefficients followed by a run of one of a predefined set of sequences, each sequence in the set being a sequence of l's and 0's starting with a coefficient of amplitude 1. In another embodiment, a first mode is defined for amplitude 1 wherein an event is a sequence of 0's and 1's of the length of the series of quantized coefficients. The method includes forming a codeword for each encountered mode and forming codewords for the formed events in the mode. Relatively short codewords are formed to represent events that are relatively more likely to occur.

Description

COMBINED RUNLENGTH CODING AND VARIABLE LENGTH CODING FOR VIDEO COMPRESSION
BACKGROUND
[0001 ] The present invention relates to image compression and in particular to variable length coding of quantized transform coefficients.

[0002] Variable length coding (VLC) is commonly used in digital video compression.
K~.iown transform based compression schemes, for example, divide an image into small blocks, e.g., 8x8 block, transform the blocks (interframe and/or intraframe) using a transform, e.g., the discrete cosine transform (DCT), quantize the resulting coefficients, order the quantized coefficients along a predefined path in the block to form a series of coefficients-sometimes called serializing the coefficients of the block, then statistically encode the series of coefficients using VLC to code the interframe and/or intraframe transform coefficients of each block.

[0003] Two-dimensional variable length coding (2D-VLC) is a commonly used VLC
technique. In 2D-VLC, each symbol-called an "event"-that is encoded by VLC is not a single coefficient, but combines a run of preceding zero-amplitude coefficients with a non-zero coefficient. That is, each event is a doublet that included) the run length of zero-amplitude coefficients preceding any non-zero coefficient, and 2) the amplitude of that non-zero coefficient. The events are statistically encoded using a variable length code such that the most frequently occurring event is encoded with the shortest codeword, and the least frequently encoded events are encoded with the longest codeword.

[0004] 2D-VLC is used, for example, in common coding schemes such as ISO/IEC
JPEG, MPEG1, MPEG2, and MPEG4 and ITU H.261, H.262, and H.263. 2D-VLC is currently being considered for use in MPEG4-partl0 and H.264.

[0005] With the advance of integrated circuit technology, computational power and memory are becoming more available. It is therefore becoming feasible to implement variable length coding schemes that provide for more compression than conventional two-dimensional VLC.

[0006] Because of the widespread use of image coding, many patents have been issued on different forms of VLC. U.S. Patent 4,698,672 issued October 6, 198'7 to Wen-hsiung Chen, one of the inventors of the present invention, for example described one form of a two-dimensional variable length coding method.

[0007] Implementing 2D-VLC typically includes modifying the method, for example to reduce the complexity of a hardware implementation. According to one modification, only a subset of the events is encoded using a VLC. Each of a set of less frequently occurring events is encoded with a relatively long, fixed-length codeword, and the other, relatively more frequent events are each encoded with a variable length codeword. An escape code at the beginning of the codeword is used to indicate the fixed-length codewords of the relatively infrequent events.

[0008] FIG. 1 shows how a table lookup may be used to implement a 2D-VLC
scheme.
Prior to the table look-up, the runlength of zero amplitudes preceding any non-zero amplitude and the non-zero amplitude are determined. The table look-up uses a 2D table for those likely events encoded using variable length encoding. The escape code fixed length codes are used for a set of relatively less likely-to-occur combinations.

[0009] In typical 2D-VLC implementations, a short end of block (EOB) code indicates the last non-zero coefficient in the block has been encountered, i.e., that the remaining coefficients in the block are all zero.

[0010] Extensions and variations to the common 2D-VLC method are known. For example, the ITU H.263 compression standard defines one such variation sometimes called three-dimensional VLC (3D-VLC). See PCT patent publication WO 9318616 published September 16, 1993 titled PICTURE DATA ENCODING METHOD (or its U.S. equivalent, U.S. Patent 5,579,413 titled PICTURE DATA ENCODING METHOD) and also the ITU-T H.263 standard. In 3D-VLC, each symbol ("event") is a triplet (LAST, RUN, LEVEL) that includes: LAST, a binary flag that indicates whether or not the current non-zero amplitude-value is the last non-zero coefficient in the block, RUN, the runlength of zero-value coefficients that precede the current non-zero amplitude, i.e., the number of zeroes since the last non-zero coefficient-amplitude, and LEVEL, the current non-zero coefficient-amplitude value. Thus, there is no need for a separate EOB
codeword; whether or not the non-zero coefficient is the last one is incorporated into the event. U.S. Patent 5,579,413 is incorporated herein by reference.
[0011 ] FIG. 2 shows how a table lookup may be used to implement 3D-VLC.

[0012] The existence of patents on 2D-VLC and extensions thereof, and the resulting patent disputes has made a search for alternate VLC methods important.
[0013] Thus there is a need for an efficient VLC scheme that can be used as a replacement of known 2D-VLC. There further is a need for an efficient VLC
scheme that can provide better compression performance than known 2D-VLC techniques.
SUMMARY
[0014] Described herein is a method and an apparatus to encode a series of quantized values that typically are coefficients of a transform of a block of image data after ordering. A first subset of values is defined that includes the most common amplitude and at least one other amplitude. A second subset of values includes amplitudes not in the first subset. In one embodiment, the transform is such that 0 is the most likely coefficient amplitude and 1 is the next most likely coefficient amplitude. In such a case, the first subset contains values 0 and 1.
[0015] The method includes forming modes for each encountered amplitude that is not the most common amplitude, and forming events for the modes in the mode modified series formed by setting to the most common amplitude all amplitudes other than the amplitude defining the mode. An event for modes for amplitudes of the second subset is a run of none or more consecutive coefficients having the most common amplitude followed by an amplitude other than the most common in the mode modified series.
[0016] In one embodiment, a first mode is defined for amplitudes of the first subset wherein an event in the mode modified series is a run of none or more coefficients of the most common amplitude followed by the coefficient of an amplitude from the first set of values.
[0017] In another embodiment, a first mode is defined for amplitudes of the first set of values wherein an event is a run in the mode modified series of none or more coefficients of the most common amplitude followed by a run of one of a predefined set of sequences, each sequence in the set being a sequence of amplitudes of the first set of values starting with a coefficient other than the most common amplitude.
[0018] In another embodiment, a first mode is defined for amplitudes of the first set of values wherein an event is a sequence of amplitudes of the first set, the sequence having the length of the series of quantized coefficients.

[0019] The method includes forming a codeword for each encountered mode and forming codewords for the formed events in the mode. Relatively short codewords are formed to represent events that are relatively more likely to occur.
[0020] Also described herein are a method and an apparatus to decode the encoded bitstream to produce a set of quantized transform coefficients.
[0021 ] Other embodiments, features, aspects, and advantages will be apparent from the description below.
BRIEF DESCRIPTION OF THE DRAWINGS
[0022] FIG. 1 shows how a prior art 2-D VLC method may be implemented by a table lookup.
[0023] FIG. 2 shows how a common prior art variation of 2-D VLC called 3D-VLC
may be implemented by a table lookup.
[0024] FIG. 3 shows a simplified block diagram of one embodiment of an apparatus to code a series of quantized transform coefficients.
[0025] FIG. 4 shows a simplified block diagram of one embodiment of a coding apparatus that encodes using VLC.
[0026] FIG. 5 shows a simplified block diagram of an embodiment of a decoding apparatus.
DETAILED DESCRIPTION
Conventional two-dimensional variable length coding [0027] Variable length coding (VLC) encodes a series of digital values, e.g., of a series of coefficients resulting by transforming a two-dimensional block of an image (interframe or intraframe) and quantizing the transform assigning variable length codes, e.g., Huffman codes or arithmetic codes, according to the statistics of the coefficients so that short codewords are assigned to relatively likely events and longer codewords to less likely events. The statistics may be lcnown or assumed a-priori, or collected on-the fly.
[0028] In 2D-VLC, the events for which codewords are formed include both the quantized coefficient-amplitude and the runlengths of the most likely-to-occur coefficients-e.g., zero-amplitude coefficients-that precedes any less than most likely-to-occur coefficient-e.g., a non-zero-amplitude coefficient when the coefficients are ordered along a specified two-dimensional path, e.g., along a zig-zag path for an 8x8 coefficient block or a 4x4 coefficient block to form a series of coefficients.
[0029] In the discussion that follows, it is assumed that zero-amplitude is the most likely-to-occur, e.g., the most frequently encountered amplitude, and that 1 is the next most likely-to-occur -next most common-amplitude. The methods described herein, however can work for other orders of likelihood, so that the most likely amplitude-the first amplitude-need not be 0, and the next most likely amplitude-the second amplitude-need not be 1. Furthermore, the discussion assumes that the likelihood-of-occurrence is indicated by the frequency of occurrence. Of course, the relative probability of occurrence need not correspond to the actual frequency of occurrence that occurs in any particular transformed image block or collection of transformed image blocks.
[0030] The occurrence information of the quantities being tracked may be tabulated in a two-dimensional table as shown in Table 1 below, where Sij, i, j = 1, 2, ...
is the number of coefficients that both have amplitude i and are preceded by j consecutive zero amplitude coefficients.
Table l: Statistics collected for 2D-VLC
Coeff. Runlength coefficients of zero valued Ampl. 0 2 ~ 5 1 1 . 1 ~ 1 ' I 3 ' 1 ' 6 I ...
1 . 4 . 1 ; . ~ ~

__ _ _ _ _ _ _ _ _ 1 S1o i__-S11 __-S12;___-S13i____S14 ___t___ _'____S16 __ ___' __ __ s15 I
. , __ A 1 1 . , 1 ___ __ ,___ ,__ __;__ ,___ ______ ~ __ __ _ ___________ 2 S2p S21 ~ S22 ~ S23 ~ ~ S25 ~ S26 ~ ...

I
____________.._____________________________________.._____________ _______________________..____________ I I

3 S3~ ~ S31 S32 ~ S33 ~ ~ S35 ~ S36 ...
. J S34 L J
I A A
L

_______________________________________________________ ___________________________ 4 _ I I ____________ S4~ ' S41 s42 ; s43 ; ; s45 ; S46 . ...
1 __ J__ s44 __ J
L _________________ L A
__ L

____________- - _ __ _________________________ s _____________ _ _ ____________ S50 ~ S51 _ I S54 _ S55 ~ S56 ~ ...
~ S52 ~ S53 L L J
I _ _____ _ A
L _ J

_________________________________ ______ __________________________ 6 _ _ _____ _ 1 ____________ S6o 1 _ ; s63 ; ; S65 1 . S61 S62 s64 I ' S66 ' ...
_ I ___________J____________ ____L___________J________ L L________ 1 __________._ ____________ ____ 7 _ I I ____________ Sip ; S~1 ~2 S~3 S~4 ~5 S~6 ! ...
_ ; ___________J____________L________ ___L___________J____________1_____ __________i_ _ L _ _ ____________ 1 __ 8 _ 1 . Sg2 ; Sg3 Sg4 ~ Sg5 _ Sgp ~ Sgl ___________; ____________~____________ ____________' .;____________i ~ Sg6 ; ...
4___________ a____________~____________ ._ I . . 1 9 S9p ~ S91 S92 ~ S~3 ~ ~ Sg5 ~ S96 ~ ...
____________~ ___________~____________Sg4 ____;..___________a____________:____________ ~____________;_ ~________ ; 1 1 .. .. ; ... .. ... ... .. I ... ...
, [0031 ] In 2D-VLC, the symbol ("event") is assigned a variable length code according to the frequency of occurrence (the number Sl~) and the particular VLC scheme-e.g., Huffman code or arithmetic code-being used. The most frequently occurring event, i.e., the largest Sij-typically Slp in transform coding-is assigned the shortest code while least occurred combination (sij) is assigned a code with the highest number of bits. The resulting 2D-VLC may then be represented by Table 2, where Cl~ is the VLC
assigned to the event (i,j): the combination of the coefficient having amplitude i that is preceded by j consecutive zero amplitude coefficients.
Table 2: 2D-VLC
Coeff. Runlength of zero valued coefficients Ampl. 0 ' 1 2 3 4 ; 5 ( . ...
' ' ' ' ' ' 1 Cl C11 C12 i C13 ; C14 ; C15 , C16 ; ...
' p ~
.

__;____________ ___________~____________;__________ __,____________~____________ 2 _ C22 ' C23 ; ~ C25 ____________ C20 ; C24 ; C26 ; ...

;

____________'_ ___________~____________~__________ _____________~____________~____________ 3 C3p ~ C32 ~ C33 ~ C35 ; C36 ~ ...
C31 __________=~ ~ ~
; C34 ~
____________~_ ~____________~_____ _______ ____________ 4 C4p ~ C41 C42 ~ C43 ~ C44 _______________________ ____________~ ___________~____________~______ ~ C45 ~ C46 ~ ...
_ ___________~ ~ ;
~

; ____ __ ____________ Csp ~ Cg Cg2 ; C53 ~ C54 _______________________ ~ l ; t ~ Csg ; . C56 ~ .
; ; . .
. a ~

________________________________________________________ ____________ 6 _ _ . C62 ~ C63 __C64 ._______________________ C6p ; C61 _ ~ _ ~ _ ' C65 ~ C66 ~ . .
__ _ . __ ____ , - , .
' _ ' __ ___ _ __ _- __ _ __ _ 7 C7p C71 C72 1 i C74 _; ___~75i____C76 i C73 . __ . ' ' _________________' 1 ' ____________4_I_ ' ____ . ____________N___________..'____________i____________ 8 Cgp ~ Cgl C82 ____________C84 . Cgs ' Cg6 ; ...
____________.______________________~ C83 ' __________:__ ~
._ ,____________:__ _____________________i____________ 9 ' . ' ;
gp ~ 91 92 C93 94 ~ C95 ' C96 ~ ...
_____ _______;____________,___________;____________;____________~___________;________ ____t____________ ...
... , ... ... , ... ... ... ... , ...

[0032] In practice, the statistics (likelihoods of occurrence of the events) are determined a-priori-e.g., by experiments on likely sets of images-to form fixed VLC
tables.
Adaptive coding also may be used in which the encoding depends on statistics such as histograms collected from actual data.
[0033] In a typical transform encoding scheme that uses 2D-VLC, the transform blocks are identified as interframe (with motion compensated or not) or intraframe, and the series of quantized coefficients within each transformed block are coded using the 2D-VLC
shown in Table 2 when ordered along a two-dimensional path.
[0034] In a typical implementation, a sign bit is added to identify whether the non-zero amplitude is positive or negative. Furthermore, some relatively non-frequent combinations are encoded by a fixed length code representing the combination of the runlength of zeroes and the amplitude following the runlength. An escape code identifying that a fixed length code is being used also is used at the beginning (or end) of each such fixed length codeword.
[0035] The Chen et al. 2D-VLC method of U.S. Patent 4,698,672 selects the variable length codes Cl~ as follows: each runlength of zeroes followed by an amplitude 1 is encoded by a first type of variable length code for the runlength. The amplitude (1) that follows the runlength of zero amplitudes need not be encoded because any codeword encoded according to the first type of variable length code can be assumed at the decoder to have amplitude 1. Each runlength of zeroes followed by a non-zero amplitude other than 1 is encoded by a second type of variable length code for the runlength followed by a code for the amplitude.
[0036] U.S. Patent 4,698,672 also describes the following extension. Suppose the series of signals, e.g., the series of transform coefficients to be encoded take on at least n+3 possible values, e.g., 0, 1, ..., n, and at least two other values, e.g., n+1, n+2, . ... The method uses n different types of mnlength codes-a first, a second, ,..., an n'th type-to respectively represent runs of none or zeroes followed by a 1, a 2, ..., an n, respectively.
The method further uses a different type of runlength code-i.e., different from the first, second, ..., n'th type-to encode runs of zeroes followed by any one of the other values.
Amplitude encoding is also to encode the amplitude, i.e., the other value that follows the run of zeroes. When one of the first, second, ..., n'th type of code is used, the type of runlength code identifies which amplitude follows the run of zeroes. In such a case, there is no need to further amplitude encode the amplitude that follows the 0's.
[0037] FIG. 3 shows the length of each codeword for a common 2-D VLC used for the H.261 standard. The MPEG-1 and MPEG-2 2D-VLC tables are similar to that for H.261.
Only the 113 most likely-to-occur symbols (runlength, amplitude-level) are represented.
MPEG-1 and MPEG-2 use similar tables. In MPEG-1, the non-common events are encoded by 20 bits or 28 bits. The escape symbol is 6 bits, followed by a code for the runlength of zeroes followed by the amplitude level code. The 20-bit escape codewords describe levels in the range [-127, +127] while the 28-bit double escape codewords have a range of [-255, +255].
MPEG-2 increases the span to the full dynamic range of quantized inverse DCT
coefficients, [-2047, +2047] and simplifies the escape mechanism with a single representation for such an event. The total length of the MPEG-2 escape codeword is 24 bits (6-bit escape code followed by a 6-bit runlength value and 12-bit amplitude level code).
Set of 1-D variable length coding tables [0038] One aspect of the present invention replaces the 2-D variable length coding table such as Table 2 by a set of 1-D variable length coding tables, each defining a "mode." In addition, another variable length coding table identifies which mode. i.e., which 1-D table of the set of 1-D tables to use. Thus, a codeword includes a first code to identify which 1-D table, i.e., which mode, is used, and a second code to identify the set of events for that mode, i.e., the events that use that 1-D table.
[0039] Consider all the possible coefficients, and consider separately each coefficient-amplitude, i.e., with all other amplitudes 0. By the "mode-modified series" is meant the ordered series of coefficients with all amplitudes other than the amplitude defining the mode set to zero. Thus, for example, for the following series of 16 coefficients that, for example, may be the coefficients of the transform of a 4 x 4 block:
O1-103-10100100000, The amplitudes are 0110310100100000.
[0040] One considers, for amplitudes of 1, the mode-modified series of coefficients:
0110010100100000;
for amplitudes of 2, the mode-modified series of coefficients:
0000000000000000;
and for amplitudes of 3, the mode-modified series of coefficients:
0000300000000000;
and so forth.
[0041] The VLC method is typically applied to transform coefficients of blocks, e.g., 4x4 or 8x8 blocks that transformed into a set of transform coefficients for the block. These blocks may be after motion compensation. The transform coefficients are quantized and the quantized block transform coefficients are ordered by scanning the two-dimensional block along a predetermined path, e.g., a zig-zag path. The ordering forms a series of coefficients for the block.

[0042] In one embodiment, the statistics of the number of contiguous zero-valued coefficients preceding a non-zero amplitude are collected for events for each such sequence having only 0's and coefficients that have a particular amplitude (all other amplitudes considered as 0). Such statistics may be written as a series of 1-D
table, one such 1-D table per non-zero amplitude. Such a set of 1-D statistics tables is shown below as Tables 3-1, 3-2, 3-3, ...:
Table 3-1: Statistics collected for runs ending with 1 (all others 0) Runlength of zero valued coefficients 0 1 . 2 3 ~ 4 1 5 6 . ...
.____ __1____________J____________ __ ____ ___ I
iSo I 1S1 I 1S2 I 1s3_ _1 ___1S4 I iSs __ ___iS6 I ,..
I t L 1 p ; I Y

Table 3-2: Statistics collected for runs ending with 2 (all others 0) Runlength of zero valued coefficients 0 1 2 3 ~ 4 5 6 ...
____ _ ;____________;____________; ___________ ____________ ;____________;
___________ .____________.
2'S0 ; 2S1 I 2S2 ; ZS3 ' 2S4 ; ZS5 ; 2S( ~ ...

Table 3-3: Statistics collected for runs ending with 3 (all others 0) Runlength of zero valued coefficients 1 . , , 0 1 2 I 3 ; 4 ~ 5 a 6 ; ...
__ 3So____'____3Si ___ i____3S~ ___i ___ 3S3____' ___3S4 __ ___ 1 1 ;
and so forth, where iSj is the number of runs of length j, i.e., for a run of j zero-amplitude coefficients ending in a coefficient of amplitude i, i=1, 2, 3, ...., with all amplitudes not equal to i considered 0.
[0043] Note that the statistics in these tables are different from those in the different rows on Table 1. For example, S3g in Table 1 is for sequences that have three consecutive zero amplitudes that are followed by amplitude 3 starting from any previous non-zero amplitude coefficient, while in Table 3-3, 3S3 is for sequences that have three consecutive zero amplitudes that are followed by amplitude 3 starting from the previous amplitude-three coefficient.

[0044] Based on these statistics, or on a set of a-priori likelihoods, a series of 1-D coding tables may be constructed as follows:
Table 4-1: 1-D VLC table for runs ending with 1 (all others 0) Runlength of zero valued coefficients 0 ~ 1 2 ; 3 . 4 . 5 . 6 . , . . , , ...
____________.____________a____________.____________.____________.,____________~
____________~____________.
1C0 ~ 1C1 ~ 1C2 ~ 1Cg ~ lCq, ~ 1C5 ~ 1C6 ; ...
Table 4-2: 1-D VLC table for runs ending with 3 (all others 0) Runlength of zero valued coefficients 0 ; 1 ; 2 ; 3 ; 4 ; 5 ; 6 ; ...
____________I____________,____________1 ____________, ___________ ,____________1 ____________.____________.
L ! L

2C ; 2C ; 2C ; 2C . 2C . 2C . 2C ~ ...
0 ; 1 ; 2 ; 3 ; 4 ; 5 ; 6 ;
Table 4-3: 1-D VLC table for runs ending with 3 (all others 0) Runlength of zero valued coefficients 0 1 ' 2 ~ 3 4 ; 5 6 . ...
A ~, ________________________ ____________L___________. J t 3C1 i 3C2 ; 3~3 ; ___3C~___ i____3~5 ___; ___ 3C6____1 ____:.:____.
and so forth, where iCj is the codeword in Table 4-i, i=1, 2, 3, ..., for a run of length j, i.e., for a run of j zero-amplitude coefficients ending in a coefficient of amplitude i (all other amplitudes considered as 0-amplitudes). The most frequently occurred statistics is assigned the shortest number of bits and the least occurred statistics is assigned the longest number of bits.
[0045] In addition, a set of statistics may be gathered for occurrences of events in each of the Tables. Such a set of statistics may be described by Table 5 as follows:
Table 5: Statistics for each of Tables 3-l, 3-2, 3-3, ...
Amplitude table ends with 1 ' 2 . 3 . 4 . 5 . 6 ; 7 . ...
___________.____________~____________.____________.____________.____________.__ __________; ___________.
1sT . 2ST ; 3ST . 4ST ; SST . 6ST ; SST ...

where 1ST is the numbers of events in the mode-modified series for the mode defined by amplitude i, i.e., iST the number of runs of zero of any length that end in a coefficient of amplitude i when all other amplitudes are considered as 0. Note that TSi is the total number of entries in Table 3-i, i=1, 2, 3, .. ..
[0046] Based on these statistics, or on assumed likelihoods for a run of zeroes followed by a coefficient being ending in a particular amplitude (all other amplitudes 0), a 1-D
VLC table is constructed as Table 6:
Table 6: Code for each of Tables 3-1, 3-2, 3-3, ...
Amplitude table ends with , . , . , , 1 ; 2 ~ 3 ; 4 ; 5 ; 6 ; 7 .___ ~ ___~____________f___________ i __________5____________Y___ i __________.
3CT ~ 4CT ~__ SCT 6CT 1 7CT __-i - ...
, ~ . , , where iCT is the codeword for using mode i, i.e., for using Table 4-i, i=1, 2, 3, .... Again, the most frequently occurred statistics (usually 1CT in transform coding) is assigned the shortest codeword and the least occurred table is assigned the longest codeword.
[0047] In one embodiment the coding tables, e.g., Tables 4-1, 4-2, ... ~ and Table 6 are fixed, determined according to statistics are determined a-priori, e.g., by experiments on likely sets of images. In an alternate embodiment, the coding tables are changeable according to data collected on the fly, e.g., according to statistics such as histograms collected on the fly, e.g., in the form of tables such as Tables 3-l, 3-2, 3-3, ... and so forth.
[0048] To encode the following sequence of 16 transform coefficients, e.g., of the transform of a 4 x 4 block:
O1-103-10100100000, the signs are ignored (see later for encoding the sign), the amplitude 1's are encoded using Table 4-1 as if there were no amplitudes other than 0's and 1's, that is, for the mode-modified series for the mode defined by amplitude 1:
0110010100100000.
[0049] The amplitude 3's are encoded using Table 4-3 as if there were no amplitudes other than 0's and 3's, i.e., for the mode-modified series for the mode defined by amplitude 3:
0000300000000000.
[0050] A code from Table 6 precedes each set of codes for each of the mode-modified series. The resulting code sequence (ignoring sign) is:
1CT 1C1 1C~ 1C2 1C1 1C2 3CT 3Cq, EOB
[0051 ] where EOB is a codeword that indicates the end of block. In one embodiment, a single end of block (EOB) codeword is used to indicate no more codes for the sequence;
no EOB is placed at the end of each sequence if the next codeword is a codeword from the mode table (Table 6). In another embodiment, the sequence of codewords for each mode end in an EOB.
[0052] Note that the method may be extended by one dimension in the same manner as traditional 2-D VLC can be extended to what is called 3-D VLC such~that there is no need for an EOB codeword. Each table, in addition to having the runlength, includes binary flag LAST, that indicates whether or not that coefficient is the last coefficient in the block of that amplitude. See above-referenced PCT patent publication WO 9318616, incorporated herein by reference.
[0053] In this discussion, it is assumed that 0-amplitude is the most likely-to-occur, e.g., the most frequently encountered amplitude, and that 1 is the next most likely-to-occur-next most common-amplitude. This corresponds to experience that events that end in 1 occur more frequently than events ending with amplitude 2 or higher, especially for interframe coded blocks, e.g., transformed blocks of motion compensated prediction residuals. This is particularly true for small blocks, e.g., 4 x 4 blocks that occur in the proposed H.264 standard.
[0054] Thus, with the above described embodiment using a series-of 1-D tables, we expect the runs to be longer than with 2-D VLC; the run in each event in any mode, i.e., using any particular 1-D table is a run from the previous identical coefficient, while in 2-D VLC, the run is from the previous non-zero-amplitude coefficient.
Nevertheless, because the codes for each 1-D table (Table 4-I) is expected to be shorter than 2-D table (Table 2) and because most codes from the amplitude-1 table, and the code indicating such a table is likely to be relatively short, we expect the complete code for a sequence of amplitudes to be competitive with that using 2-D-VLC.

[0055] In one embodiment, some relatively non-frequent events are encoded by a fixed length code representing the combination of the runlength of zeroes and the amplitude following the runlength. An escape code identifying that a fixed length code is being used also is used at the beginning of each such fixed length codeword. In one implementation, a single representation is used for the less likely-to-occur events that are encoded by a fixed length codeword.
A variable length coding table of likely events and a further set of 1-D
variable length coding tables [0056] Another embodiment produces efficient codes in the case that events that are a sequence of none or more zero-amplitude coefficients followed by patterns of 0-and 1-amplitudes. Such events are assumed but not required to be more likely to occur than events that are a sequence of none or more zero-amplitude coefficients followed by an amplitude larger than 1. Such a situation may be applicable, for example, for small interframe coded blocks, say 8 x 8 or 4 x 4 blocks of interframe transform coefficients after motion compensation.
[0057] According to such an embodiment, a first subset of amplitudes is defined, including the most common amplitude and one or more of the next most common amplitudes. The other amplitudes are defined as a second subset of amplitudes.
In one implementation, the first subset of amplitudes consists of 0 and 1, the two most likely-to-occur amplitudes, and the second subset consists of those amplitudes that are larger than 1. A set of tables is defined. The first table is for runs of the most likely-to-occur amplitude followed by one of the predefined set of sequences, where of the predefined set of sequences is a sequence of at least one coefficient from the first subset of amplitudes, starting with an amplitude other than the most likely-to-occur. Because events are defined by runs of none or more coefficients having the most likely to occur amplitudes, each of the sequences in the predefined set typically also ends with an amplitude other than the most likely-to-occur. All other amplitudes are set to the most likely-to-occur. Typically, the first subset consists of 0 and 1, and the sequences in the predefined set are sequences of only 0's and 1. Thus a first mode is defined by amplitude 1, and the events are runs of none or more zero-amplitudes followed by one or more repetitions of a sequence of the predefined set in the mode-modified series. The other tables are each ,for a mode defined by one of the amplitude of the second subset of amplitudes, e.g., an amplitude greater than 1. An event is a run of none or more of the most likely-to-occur amplitude followed by a single amplitude of the second subset of amplitudes, with all other amplitudes set to the most likely-to-occur, i.e., for the mode-modified series for the mode defined by the single amplitude. There are further tables, one for runs of none or more 0-amplitides followed by each of 2, 3, 4, and so forth. These tables for the same events as shown above in Tables 3-2, 3-3, ...
[0058] Once the set of sequences is defined, in one embodiment, statistics may be gathered for the symbols (events) that include a run of 0 amplitudes followed by a run of one of the set of sequences of the predefined set, all other amplitudes set to 0. Table 7-1 shows one such set of statistics.
Table 7-1:
Statistics for coeff.
ampl.
patterns preceded by zero coef amplitudes Coeff. Runlength zero valued Ampl. of coefficients Pattern . ' ; .
0 . 1 ; 2 ; 3 ; ...

1 S 1-1-0 . S 1-1-1 S 1-1-2 S 1-1-3 _______ ____ . ; ___ _______ ____ ____ ii S1-2_~ S~=2-1_____i_____S1-2-2_____i_____ _____Si-2 3_____ ______ ______ __;__________________ __________________;.___________________;.______...
;

______ 111 S 1-3-0 ; S 1-3-1 S 1-3-2 S 1-3-3 ; ;

___________________________________;___________________;___________________,.__ _________________;______...______ ...
. ;

_________________.. ... , ... ; ... , ...
__ u___________________a___________________a___________________a_______________ ____111111________S1=6_0________S1=6_1 . S1-6-3 ...
S1=6_2 101 5101=1 ____ ____ S1-1-3 ;
0 ; S1-1-1 ____ ______ ___ ...
______ _ y___________________~___________________Y___________________y._______________ __ 1001 S 1001-1-0 ; S 1001-1-1S 1001-1-2S 1001-1-3 ; . ...

__ _____ __ __ __ ___ _ 1101 S 1101-1-0 __ ___~___ S 1101-1-3 ...
______ S 1101-1-1 __a_ __________________a_______________ ______1011__5101_1_ __5101_1_ 51011-1-2 51011-1-3 ...
_____ 1=0__ 1=1 10001 510001-1-0 __ _ _ 510001-1-3 . ...
. 510001-1-1510001-1-2 .

_____ __ ~___________________1___________________T_ _______________ ____ S 11001-1 ' S 11001 1 ; S 1100 - - 1-1-2 ; S 11001-1-3 ; ...
-10101 510101-1-0 ; 510101-1-1510101-1-2510101-1-3 ...

_____ __ ~___________________~___________________i___________________a_______________ 10011 S 10011-1-0. S 10011 .
.

- ...

_____11101_511101=1 _511101_1 511101-1-2511101-1-3 ...
____ _0_ _1 i .
;

__ _________________;___________________r_______________ _____10111_____510111=1_0___ 510111_1_2510111=1=3 . _510111_1=1' .
i ___ _ _ _ 101101 S 101-2-0 ; S 101-2-1 S 101-2-2 _ ; ; _____ _____ S 101-2-3 . ..

[0059] The quantity Si j-k in this case is the statistic for a run of k 0-amplitudess to be followed by j repeats of the sequence i. Therefore, S 101-3-4 is the sequence of coefficient-amplitudes 0000101101 which is four 0's followed by two repeats of the sequence 101.
[0060] A variable length code is then derived for each of the Si-j-k to provide a VLC
table as shown in Table 8-1.

Table 8-1:
VLC
for coeff.
ampl.
patterns preceded by zeroes Coeff. Runlength of zero valued coefficients Ampl. 0 1 ~ 2 3 ' ' ' Pattern ; ; .

1 C1-1-0 C1-1-1 C1-1-2 C1-1-3 ...

______ __ J__________________J____________________J___________________J______________ ______1l___________C1=2=~____ ' ___C1=~=1______ Cl 01-2=3 ~ _2_2 ~
~
...

____ ____ 111 C1-3-0 C1-3-1 . C1-3-2 ____ ____ ____ . C1-3-3 . ...

___________ _ _ __________________.~__________________~____________________~___________________ T_____________ _ __ 1 1 I 1 .. 1 ...
..

... ...
___ i I _____ ______ ______ 111111 l-6-0 ; Cl-6-1 0l-6-2 C1-6-3 , ...
__________________ ____________________~___________________,___ ;
_ _ ___________________J______________ 101 0101-1-0 0101-1-1 ______________0101-1-3 ...
____ 0101-1-2 __.______________ 1001 01001-1-0 01001-1-1 01001-1-2i 01001-1-3 ...
i ____ __ 1__________________,~__._________________~___________________Y______________ _____ __ 01101=1 _l_ 01101 ____ 1101_______ . _1_ 2 01101=1=0__ 01101=1 _3 .
___ _ _ _ 1011 01011-1-0 ; 01011-1-1 ___ __ _____ ; 01011-1-2____ ; 01011-1-3 ; ...

____ __ ,__________________,____________________J___________________ 10001 010001 010001 ;______________ ~

- ; - 010001-1-3 - - -2 ...
-____11001_____ 011001 _l_ 011001 .
01_1001-1-0_ 1 . _l_2 011001_1 _3 ___ _ _ _ 10101 010101 010101=1-1 010101 _ _____ 1=0 1-2 ____ 010101-1-3 . ...

___ _ __~__ __________________,___________________;.______________ ____ _ ~ 010011-1-1010011-1-2; 010011-1-3 ; ...
10011 _ ;

____ __ ;__ __________________;___________________ ;______________ 11101 011101-1-0 011101-1-1 011101-1-2; 011101-1-3 ; ...
__________________ ; _______ J
____________________ J____________________ __ J

_ ___________________J______________ ___ 10111_____ 010111=1_ ________ 010111_1 _3 010111=1_0_ 1 _ 010111 .
_1=2_ ___ _ 101101 0101-2.-0 . 0101-2-1 0101-2-2 _ . _____ ____ . 0101-3-3 . . .
.

[0061 ] The other tables are identical to Tables 3-2, 3-3, etc., and the coding tables are the same as Tables 4-2, 4-3, etc.
[0062] Different versions can include more or fewer sequences in the set of sequences.
One embodiment includes every sequence that starts and begins with amplitude 1.
Another includes just those shown in Table. 8-1. Furthermore, while one version has some repetitions of each sequence, another version has only one instance of each of the sequences. Yet another version has repetitions of some sequences, e.g., of single amplitude 1's, but not of some other sequences. For example, one version includes only up to six consecutive amplitude-1's, and only a sequence of two 101's, as shown in Table. 8-1.
[0063] In addition, a Table of statistics similar to Table 5 may be collected, except that the first column are the statistics for the events that are in Table 8-1.

Table 9: Statistics for each of Tables 7-l, 3-2, 3-3, ...
Amplitude table ends with 'i 1 . 2 3 ~ 4 5 . 6 7 ...
___________ ____________a____________~___________.____________;____________~____________,__ __________ a I 1ST . 2ST ' 3ST ; 4ST SST ; 6ST ; SST ~ ...
where 1ST is the number of runs of Table 7-l and iST, i>1 is the total number of entries in Table 3-i, i=2, 3, 4, ....
[0064] Based on these statistics, or on assumed likelihoods for a run of zeroes followed by a coefficient being ending in a particular sequence or repetition of a sequence of the predefined set, or an amplitude > 1, (all other amplitudes 0), a 1-D VLC table is constructed as Table 10:
Table 10: Code for each of Tables 7-l, 3-2, 3-3, ...
Amplitude table ends with 1 ; 3 ; 3 ; 4 . S ' 6 . 7 ; ...
___ iCT ___, ___2CT___ i___ 3CT ___i ___4CT ___i ___SCT___ i___ 6CT ___;
___7CT ___; ____:.: ____ Where 1CT is the code for using Table 8-l, and iCT , i>1, is the code for using Table 4-i, i=2, 3, 4, .... Again, the most frequently occurred statistics (usually 1CT in transform coding) is assigned the shortest number of bits and the least occurred table is assigned the longest number of bits.
[0065] Again, one embodiment uses fixed tables, constructed, for example, from a-priori statistics. Such statistics may be obtained from experiments performed to determine the set of most likely sequences and the relative likelihood of a run of 0's following one or more repeats of such predefined sequences, all other coefficients 0. In another embodiment, the tables are changeable according to actual data.
[0066] To use the coding tables, in one version, after the transform blocks are identified as interframe (including motion compensated interframe) or intraframe mode, the series of quantized transform coefficients within the block are coded according to the above table.
[0067] As an example, to encode the following sequence of 16 transform coefficients, e.g., of the transform of a 4 x 4 block:
O1-103-10100100000, the signs are ignored (see later for encoding the sign), the amplitude 1's are encoded using Table 8-1 as if there were no amplitudes other than 0's and 1's. That is, for the mode-modified series for amplitude l:
0110010100100000.
[0068] The amplitude 3's are encoded using Table 4-3 as if there were no amplitudes other than 0's and 3's, i.e., for the mode-modified series for amplitude 3:
0000300000000000.
[0069] A code from Table 10 precedes each set of codes for each of the sequences. The resulting code sequence (ignoring sign) is:
iCT C11001-1-1 C1001-1-1 3CT 3Cq. EOB
where EOB is a codeword that indicates the end of block. In one embodiment, a single end of block (EOB) codeword is used to indicate no more codes for the sequence; no EOB
is placed at the end of each sequence if the next codeword is a codeword from the mode table (Table 10). In another embodiment, the sequence of codewords for each mode end in an EOB.
[0070] In one embodiment, some relatively non-frequent events are encoded by a fixed length code.
A variable length coding table of all events with likely coefficients and a further set of i-D variable length coding tables [0071 ] The current proposed transform coding scheme MPEG-4 part 10 and H.264 transform encodes blocks of 4 by 4 coefficients. The likelihood of amplitudes larger than 1 is then much smaller than of amplitudes of 0 and 1. A sequence of 16 of 0's and 1's has 216 possible arrangements. In another embodiment, the first table, of only 0's and 1's, includes variable length codes for all possible combinations of 0's and 1's.
With the falling cost of memory, storing such a table is practical for 4 x 4 (and even larger) blocks. Such a 1-D code table is shown below as Table 11-1.

Table 11-1: VLC for coeff.
ampl.
patterns Coeff. Ampl. Pattern VLC

0000000000000001 __________ C-1 __ 0000000000000010 _ _ C-2 _ _ _ _ 0000000000000011 _ _ _ _ _ C-3 _ _ _ _ 0000000000000100 _____________ __ C-4 0000000000000101 _ __ _ _ _ _ _ C-5 _ 0000000000000110 _ _ _ _ _ _ _ _ _ _ _ _ C-6 _ 0000000000000111 _ _ _ _ _ _ _ _ _ _ C-7 _ 0000000000001000 _ C-8 _ _ _ _ 0000000000001001 _ _ _ __ __C-216-1 [0072] Table 11-1 is constructed by either gathering statistics or by a-priori assumptions or experiments on the likelihood of each sequence (all coefficients greater than 1 set to 0), and then assigning a variable length code to each sequence. A further table of codes is constructed for when to use Table 11-1, and when to use code Tables 4-2, 4-3, etc. A code table similar to Table 6 can be constructed to assign a codeword to indicate which of Tables 11-1, 4-2, 4-3, etc. to use. In this. embodiment, 1CT is the code for using Table 11-1, and 1CT , i>1, is the code for using Table 4-i, i=2, 3, 4, ....
[0073] When a mode indicates use of Table 11-l, there is no need to determine the runlength of the preceding 0's -- the method identifies the complete sequence and looks up the VLC in Table 11-1.
[0074] Consider again the example to encode the following sequence of 16 transform coefficients, e.g., of the transform of a 4 x 4 block:
01-103-10100100000, the signs are ignored (see later for encoding the sign). The amplitude 1's are encoded using Table 11-1 as if there were no amplitudes other than 0's and 1's. That is, for the mode-modified series for amplitude l:
0110010100100000.

[0075] The amplitude 3's are encoded using Table 4-3 as if there were no amplitudes other than 0's and 3's, i.e., for the mode-modified series for amplitude 3:
0000300000000000.
[0076] A code from Table 6 precedes each set of codes for each of the sequences. The resulting code sequence (ignoring sign) is:

so that there is only a single codeword, C-0110010100100000 in the mode for Table 11-1.
[0077] In one embodiment, some relatively non-frequent events are encoded by a fixed length code.
Encoding the sign information [0078] Another aspect of the invention is encoding the sign information. Each coefficient may be positive or negative. Thus, the formed codewords for a set of events are encoded with information indicative of the sign of any non-zero amplitudes in the events.
[0079] In conventional 2-D VLC, the sign information adds a bit to each codeword. In one embodiment of the present invention, each codeword from one of the sequence of 1-D
coding tables includes a sign bit to indicate the sign. That is, when the event is a run that ends with a particular amplitude, a further bit is included in the code sequence for use of each 1-D coding table to indicate the sign bit.
[0080] Referring to the embodiment that uses a set of 1-D coding tables, consider the example of encoding the following sequence of 16 transform coefficients:
O1-103-10100100000.
the amplitude 1's are encoded using Table 4-1 as if there were no amplitudes other than 0's and 1's. That is, for the mode-modified series for amplitude l:
Ol-100-10100100000.
[0081 ] The amplitude 3's are encoded using Table 4-3 as if there were no amplitudes other than 0's and 3's, i.e., for the mode-modified series for amplitude 1:
0000300000000000.
[0082] Let s+ refer to the sign bit for a positive coefficient and let s_ refer to the code bit for a negative coefficient. A code from Table 6 precedes each set of codes for each of the sequences. The resulting code sequence is:

1CT lCls+ iCp S_ 1C2 S_ 1C1 s+ 1C2 s+ 3CT 3C4 S+ EOB
where EOB is a codeword that indicates the end of block.
[0083] This is also applicable to each of the 1-D tables for the embodiments that use a variable length coding table of some common events with likely coefficients and a further set of 1-D variable length coding tables, and for the embodiments that use a variable length coding table of all events with likely coefficients and a further set of 1-D variable length coding tables.
[0084] In a first alternate embodiment, a value of +1 is distinguished from a value of -1 for the coding table of the events that have the common amplitudes of 0 and 1 and in the mode-modified series for the mode defined by amplitude 1. Thus, Table ~-1 for a set of binary (0 or 1) amplitudes is replaced with Table 12-1 shown below of ternary values, 0, +1, and -1, where an event is a run of consecutive 0's followed by a common sequence of 0, +l, and -1, starting with either +1 or -1. Typically, the sequences also end with either +1 or -1.

Table 12-1:
VLC
for coeff.
patterns preceded by zeroes Coef Ampl. Runlength of zero valued coefficients Pattern 0 ' 1 . 2 ~ ...

+1 +1C-1-0 +1C-1-1 . +1C-1-2 ...
__ _ _______+1 --_---_+iC-2_O_____ -_ ________________________;
+1- --__-_ - ____'__________+iC-Z~i_--___________ _________________________________________________ ____,______________________________;
_ ... ; ... . ... . ...
... , , , __..______.._______________________________________ ___~__________________________-___~_________-_______________~__________ +1+1+1+1+1+1 +1C-6-0 +1C-6-1 +1C-6-2 . ...
____ ;

_________.._______________________,._______________ ___7______________________________r..______ -1 -1C_1_0 -1C_1_1 _________________.._______ -1C_1_2 ~ ...

-_--_--_ i _---____=lC-2~0_____ ____'__________~iC_2-i--___-----i_______~iC~~-2 _______;
i___-____ ________________________ ____,______________________________;___-_________________________ _________________________;
__________ ... ... , ... ...
___________...___________ ___________________________ ______________________________r_________________________~__________ -1-1-1-1-1-1 -1C-6-0 -1C-6-1 . -1C-6-2 ...

_________________________ ____,._________-_____________,_____..________________________r_________________________~_______ ___ +1-1 +1-1C_1_0 ; +1-1C_1_1 +1-1C_1_2 ...
~
;

_________________________________________________ ______________________________~__________ -1+1 -1+1C_1_0 ____ _________________________ ; -1+1C_1_1 ; -1+1C-1_2 . ...

___-_-+1+1-1--_______--~1+1-1C_1_~-_ ___ ;______-+1+1-1C-1_1-______i___-+1+1-1C_1_2-___-1 ___-_.. -_ ____-______________________-___-________..________ ___.i_-___________________-___-____;Y
1 1 ______.._______-_-____-___ ____--____ ... ... 1 ... 1 ... 1 ...
______-_____-___________- ____..________-__-_____-_-__ ~_______-_-__.______..______-___r-_____._______-__.._______r____-.____ +10+1 +lo+1C_1_0 ; +lo+iC_1_1 ; +1o+1C_1_2 ...
_ ~

_________________________ ___________________________ ~______________________________~_____..___ +10-1 +lo-1C_1_0 . +lo-1C_1_1 _________________________ ~ . +lo-1C_1_2 ...

____________..____________ ______________________________~_________________________~_____-____ -10+1 ______________________.._ ___ -1o+1C_1_2 . ...
-1o+1C_1_0 -1o+1C_1_1 ;
, ;

______-X10-1-_________-_-ul0-1C-1_~___ __ _____-~1o-1C_1_1-_______y___--_lo-1C_1-2 _____i ____:.. -_ _____-+100+1--________-+too+1C_1-o-_ ___ ;_______+loo+iC-1_1-______;___-+ZOO+1C_1-2_____i ___-:., __ _____-+140-1--________-+loo-1C_1_~ __ ____- loo-1C_1_1___- +loo-1C_1_y____;
__ _-_____; ___-_.: -_ _____--100+1-____-_too+1C_1_~__ ___;______-_loo+1C_1-1-______;___--loo+IC_1-2 ____i____.,-_ ____ __L _________________________L__________ _____-X100-1-_________-_ioo-1C_1_~ ____1_____--yloo-iC_1_1-loo-iC_1_2 ...
-_ _____ ;

_____ _____ ____ __L -______-_-___._-_-_______L-_________ +1+10+1 +1+lo+1C_1_p ; - +t+1o+iC_1_1 +1+1o+1C_1_2 ; ...

_________________________ ____,______________________________;_________________________;
-_______________________ __________ ... ...
... ... , ...
__-~l~l-10-1--______-=lo-1C_2_~___ , ____-_lo-1C_2-2 -____;
____'_______-_lo-1C-2_1-_______i___-_.. __ [0085] where quantity iC j-k in this case is the codeword for a run of k 0-amplitudes to be followed by j repeats of the sequence i, where i is a sequence that starts and ends with +1 or -l, and contains only 0,+l, and -1. Therefore, +loo-1C_3-2 would be the codeword for the event 00+100-1+100-1+100-l, which is a run of two 0-amplitides followed by three repetitions of the coefficient sequence +100-1. This assumes that the set of sequences to be encoded by this table includes three repetitions of the coefficient sequence +100-1. As described above, a subset of possible coefficient sequences that include the likely values 0, +1, and -1 is selected to be included in the table.
[0086] One embodiment selects single instances of a sequence of a predefined set of sequences-a subset of possible sequences-of coefficients that include likely values 0,+1, and -1, starting (and typically ending) in a non-zero-amplitude, while another embodiment, e.g., of Table 12-1, allows some repetitions of the sequences of the predefined set.
[0087] As an example, to encode the following sequence of 16 transform coefficients, e.g., of the transform of a 4 x 4 block:
01-103-10100100000, the amplitudes 1's are encoded using Table 12-1 as if there were no amplitudes other than 0's and 1's. That is, for the mode-modified series for amplitude l:
O1-100-10100100000.
[0088] The amplitude 3's are encoded using Table 4-3 as if there were no amplitudes other than 0's and 3's, i.e., for the mode-modified series for amplitude 3:
0000300000000000.
[0089] A code from Table 10 (assuming 1CT indicates using Table 12-1) precedes each set of codes for each of the sequences. The resulting code sequence is:
1CT +1-100-1C_1_1 +100+1C-1-1 3CT 3Cq, s+ EOB, where s+ the sign encoding for + as above for the tables of amplitude greater than l, and EOB is a codeword that indicates the end of block. This assumes that +1-100-1 is one of the sequences included in Table 12-1.
[0090] Similarly, for the embodiment that includes a variable length coding table of all events with likely coefficients and a further set of 1-D variable length coding tables, Table 11-1 can be replaced by a coding table that differentiates between +1 and -1, i.e., of all events that are sequences of 0, +1, and -1. For a sequence of N coefficients, there are 3N
1 such sequences. For a block of 4 by 4, that means 316-1 entries in the table. This is slightly over 43 million entries. With the cost of memory steadily falling, such a table is fast becoming practical.

[009'1 ] According to another aspect of the invention, the sign information is encoded in the form of information related to a change of sign. We call this differential sign encoding.
[0092] According to one embodiment of differential sign encoding, the information related to a change of sign is for any change of sign of non-zero-coefficients in the series of coefficients.
[0093] Differential sign encoding is applicable to using a series of 1-D
tables, e.g., to code Tables 4-l, 4-2, 4-3, and so forth, and for those embodiments that use a coding table of a subset of sequences of common amplitudes, or a table of all sequences of common amplitudes, to the 1-D tables for amplitudes greater than 1, e.g., code Tables 4-2, 4-3, and so forth.
[0094] According to one embodiment that uses differential sign encoding, an identifiable code is used to indicate whenever the change changes from + to -, and from -to +. One version assumes the first sign is +, while another assumes the first sign is -. Yet another starts with a code that indicates whether the first sign is + or -, and then proceeds to encode change of sign only.
[0095] One embodiment uses differential sign encoding within each mode for the mode-related series. The information related to a change of sign is for any change of sign of non-zero-coefficients in the mode related series of coefficients. Consider again the example of encoding the following sequence of 16 transform coefficients, e.g., of the transform of a 4 x 4 block:
01-103-10100100000.
Assume a method according to which the first sign is assumed +. Let s~ be a unique code that indicates change of sign. The amplitude 1's are encoded using Table 4-1 as if there were no amplitudes other than 0's and 1's. That is for the mode-modified series for amplitude 1:
O1-100-10100100000.
[0096] The amplitude 3's are encoded using Table 4-3 as if there were no amplitudes other than 0's and 3's, i.e., for the mode-modified series for amplitude 3:
0000300000000000.

[0097] A code from Table 6 precedes each set of codewords for the use of each 1-D table.
The resulting code sequence is:
1CT 1C1 sc 1Cp 1C2 sc 1C1 1C2 3CT 3C4 EOB
[0098] where EOB is a codeword that indicates the end of block, and where the change of sign codeword precedes the codeword that has the new sign. In an alternate embodiment, the change of sign codeword follows rather than precedes the codeword that has the new sign.
[0099] For the embodiment that includes an indication of the first sign, the resulting code sequence is:
1CT S+ 1C1 Sc 1Cp 1C2 Sc 1C1 1C2 3CT S+ 3C4 EOB
[00100] where s+ and s_ indicate the first coefficient is +ve and a -ve, respectively, EOB is a codeword that indicates .the end of block, and where the initial sign and change of sign codewords precede the codeword that has the sign.
[00101 ] According to another alternate embodiment, the encoding of the signs of the coefficients includes encoding the sequence of signs of all the non-zero coefficients to produce a separate sign codeword for all the signs. Only the non-zero coefficients are encoded. Such encoding can be done only according to relative positioning of the non-zero coefficients, and irrespective of the actual position of the non-zero coefficients, e.g., irrespective of the length of the run of zero-amplitude coefficients that precede a non-zero coefficient.
[00102] For the example sequence of 16 transform coefficients:
Ol-103-10100100000, the signs are +-+-++.
[00103] Because in typical transforms of residual images after motion compensation, most coefficients are zero, i.e., fewer non-zero coefficients are more likely than many non-zero coefficients, in one embodiment the coding of the sign sequence is to code 1 for + and 0 for -, preceded by a unique code that indicates the codeword follows. This provides a variable length code. As an example, the variable code word for the above sign sequence would be cs 10101 l, where cs is the code that indicates the sign codeword follows. In another embodiment, no cs code is used or needed. The sign sequence may be attached at the end of the EOB.
An apparatus for Encoding [00104] FIG. 3 shows one embodiment 300 of an apparatus for encoding according to the VLC embodiment that uses Table 6 and Tables 4-1, 4-2, ... One embodiment uses a set of 1-D lookup tables such as Tables 4-1, 4-2, ... and a lookup table such as Table 6 to include a code codeword. When applied to transform encoding, a series of quantized transform coefficients is obtained by transforming a 2D image-block, ordering the transform coefficients along a path in the block, and quantizing the coefficients. The series of coefficients is entered in an event recognizer 303 that in one embodiment includes a recognizer that recognizes non-zero amplitudes-called a mode recognizer-and generates a set of events for each such non-zero amplitudes as if all other amplitudes were zero, i.e., for each mode modified series for the mode. For each non-zero amplitude encountered, the event recognizer determines a set of runlengths of each run of 0's from the last such non-zero amplitude (or the start of the sequence) until the next non-zero amplitude encountered as if no other non-zero amplitudes existed. One version of the runlength determining uses counters that operate in parallel. In such version, a single counter 311 continues counting from the start of the sequence. Each time a previously-unencountered non-zero amplitude is encountered, the value from the main counter is transferred as the first runlength for that amplitude and a new counter is started for that amplitude. In FIG. 3, the counters for each such amplitude are shown as 313, 315,..., and annotated as cntl, cnt2,..., cntN for N different amplitudes that are found.
[00105] One embodiment of the event recognizer 303 that uses parallel counters passes sets of events in parallel to an encoder 305, one set for each non-zero amplitude encountered. FIG. 3 shows the case for N different amplitudes. Each set of events includes the amplitude (or mode) for the set and a sequence of runlengths of zeroes that are between such amplitudes (assuming all other amplitudes are 0).
[00106] The encoder 305 is coupled to a memory 307 that stores the information in the 1-D
coding tables 4-1, 4-2, ..., and also the coding table (Table 6). The encoder 305 includes a lookup device to look up the coding table to output the codeword for the mode, and then looks up the coding table for the sequence of runlengths for such mode (as if all other amplitudes are 0). One embodiment uses the runlength as an index for the codes stored in an indexed list. As described above, for some of the events, a fixed length codeword with an escape code is formed (see below). The encoder then outputs the complete sequences of codewords for each mode, each starting with the codeword for the mode. Sign information also is included.
[00107] While the embodiment shown in FIG. 3 has the memory 307 that stores the lookup table separate from the lookup device, in another embodiment, the memory is part of the lookup device of encoder 305.
[00108] FIG. 4 shows an embodiment 400 of an apparatus that implements another of the method embodiments described above, wherein one of the modes is for runs of 0-amplitudes followed by one of a set of pre-defined sequences of amplitudes 1 and 0, starting and ending with an amplitude of 1. In particular, the embodiment 400 implements a version that uses the coding tables Table 8-1, 4-2, 4-3,..., and the mode coding Table 10. In one version, the coding table 8-1 only includes single repetitions of any of the pre-defined sequences. How to modify for repetitions of any of the pre-defined set of sequences would be clear to those in the art.
[00109] As above, when applied to transform encoding, a series of quantized transform coefficients is obtained by transforming a 3D image-block, ordering the transform coefficients along a path in the block, and quantizing the coefficients. The series of coefficients is entered in an event recognizes 403 that in one embodiment includes a sequence pattern recognizes that recognizes which of a predefined set of M
sequences of amplitude-1's and 0's follows a run of none or more 0's. The event recognizes 403 further f recognizes which of non-zero amplitude greater than 1 follows a run of none or more 0's.
The event recognizes further determines the runlength of the run of 0's from the last recognized sequence of the predefined set as if all other amplitudes are 0.
Furthermore, for each non-zero amplitude other than 1 encountered, the event recognizes determines a set of runlengths of each run of 0's from the last such non-zero amplitude (or the start of the sequence) until the next non-zero amplitude encountered as if no other non-zero amplitudes existed.
[00110] One version of the runlength determining uses counters that operate in parallel. In such version, a single counter 411 continues counting from the start of the sequence. The first time one of the predefined sequences is encountered, a counter 413 is stated. Each time a previously-unencountered non-zero amplitude other than 1 is encountered, the value from the main counter is transferred as the first runlength for that amplitude and a new counter is started for that amplitude. In FIG. 3, the counters for each such amplitude are shown as 415, ..., and annotated as cnt2,..., cntN for N-1 such different amplitudes greater than 1 that are found.
[00111] Each of the predefined set of sequences has a sequence number. The event recognizer 403 is coupled to memory 407 that stores the subset of sequences.
The event recognizer outputs information to an encoder 405. The outputs for amplitudes greater than 1 are the same as for the embodiment 300 of FIG. 3, and for the sequence, the output is a mode indicator, followed by a sequence of two numbers for each recognized sequence:
the recognized sequence number and the runlength of 0's since the last recognized sequence as if all other amplitudes were 0.
[00112] The encoder 405 that in one embodiment includes a lookup device that is coupled to a memory that stores the information in Table 8-1, Tables 4-2, 403, ..., and Table 10.
In one embodiment, the same memory 407 is used to store the tables as is used to store the sequences. In another, separate memories are used.
[00113] While the embodiment shown in FIG. 4 has the memory that stores the lookup table separate from the lookup device, in another embodiment, the memory that stores the coding tables is part of the lookup device in encoder 405. The phrase "the lookup device is coupled to a memory" includes the case that the lookup device includes the memory.
[00114] As is known in the art, an end-of-block (EOB) indicator can be used with any of the encoding schemes described above to indicate the end of a block, i.e., that all remaining coefficients in the block have zero amplitude.
[00115] In alternate versions of each of the encoding schemes, whether or not the end of block is reached can be included as part of the event. In other words, an event now includes an additional "dimension" called "last" that indicated whether or not the event is the last in the block, i.e., the remaining coefficients in the block following the event axe all the most common amplitude (zero).
[00116] Note that the apparatuses described above can be modified to include all the alternatives described above for encoding the sequence of coefficients, and how to so modify the apparatuses would be clear to one in the art. Furthermore, the apparatuses described above can be modified to include all the above described above methods for encoding the sign information, and how to so modify the apparatuses would be clear to one in the art.
Decoding [00117] Another aspect of the invention is a method for decoding a bitstream encoded by any of the encoding methods described herein. For example, one aspect is a method to decode a bitstream that is coded by a method that includes modes of codewords.
In one embodiment, the mode is defined as a non-zero amplitude. A codeword defines the mode, and within each mode, a codeword is formed for each run of none or more zero-valued coefficients followed by that amplitude when all other amplitudes are defined as 0. In another version, one of the modes is for amplitude 1 coefficients, and for that mode, a codeword is formed for each run of none or more zero-valued coefficients followed by a run of one of a predefined set of sequences, with all other non-zero amplitudes that are not one considered as zero. Each sequence in the predefined set is a sequence of amplitude-1's and 0's staxting and ending with a coefficient of amplitude 1. The coding is such that relatively short codewords in the bitstream represent events that are relatively more likely to occur, and relatively long codewords represent events that are relatively less likely to occur. The decoding method includes determining the mode, and then for the mode, for each codeword, determining the event represented by each codeword, and determining the sub-series of coefficients of each determined event until all coefficients in the block are determined. The decoded modes are then combined to provide the original sequence. In one version, a codeword representing an end of block is used in the coding to indicate that all remaining coefficients in the block are 0. When the decoding method encounters the codeword representing the end of block, it assigns amplitude 0 to all remaining coefficients.
[00118] How to recognize the codewords in the bitstream and how to decode the recognized codewords given a description of the encoding scheme, e.g., a coding table, is known to those in the art. In one embodiment, the codes are selected such that no codeword is the prefix of a longer codeword. The encoding method includes recognizing a codeword bit-by-bit as more data of the input bitstream is obtained. One method uses a decoding table, which is a table that provides the events for all possible codewords. As a codeword is recognized, the event of that codeword, i.e., the sequence of coefficients is formed.

[00119] One decoding method uses a lookup table. In such an embodiment, a table is provided containing a plurality of codewords and the events for each codeword.
The determining of the event for each recognized codeword includes looking up the table.
[00120] Another aspect of the invention is an apparatus for decoding a bitstream encoded by any of the novel methods described herein. FIG. 5 shows one embodiment of the apparatus. A codeword recognizes 502 accepts the bits of a bitstream and recognizes a codeword of a set of possible codewords that each represents an event or a mode. A
decoder 504 is coupled to the codeword recognizes 502 and determines the mode or the event for the codeword recognized by the codeword recognizes. In one embodiment, the decoder 504 is a lookup device that looks up a table stored in a memory 506.
That table provides the event for at least some of the codewords of the set of codewords.
Other codewords may include an escape code, so that decoding is by other than a table lookup.
In one embodiment, the bitstream is stored in a buffer 908, and the output of the buffer is coupled to the codeword recognizes.
[00121] While in the embodiment shown in FIG. 5, the memory is shown separate from the lookup device, those in the art will understand that in other embodiments, the memory for the table is part of the lookup device, and such other embodiments are included herein.
[00122] Another aspect of the invention is a memory (such as memory 307 in FIG. 3, memory 407 in FIG. 4, and memory 506 in FIG. 5) that stores a coding data structure that provides the codeword for an event as described any one of the novel~coding methods described herein. In one embodiment, the data structure is in the form of one or more tables. For example, one memory embodiment stores a table that provides the codeword for the mode, and then includes tables for events that are for each non-zero amplitude each the length of a run of none or more zero-valued coefficients that are followed by that non-zero amplitude as if all other non-zero amplitudes are zero. There is one such table for each amplitude. In another embodiment, one of the tables defines events that are each a run of none or more zero-valued coefficients that are followed by a run of one of a predefined set of sequences, each sequence in the set being either a single coefficient of amplitude greater than one, or a sequence of 1's and 0's starting with a coefficient of amplitude 1, as if all other non-zero coefficients were zero. Relatively short codewords in the table represent events that are relatively more likely to occur, and relatively long codewords represent events that are relatively less likely to occur.

[00123] Reference throughout this specification to "one embodiment" or "an embodiment"
means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the present invention.
Thus, appearances of the phrases "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment.
Furthermore, the particular features, structures or characteristics maybe combined in any suitable manner, as would be apparent to one of ordinary skill in the art from this disclosure, in one or more embodiments.
[00124] Similarly, it should be appreciated that in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. This method of disclosure, however, is not to be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the Detailed Description are hereby expressly incorporated into this Detailed Description, with each claim standing on its own as a separate embodiment of this invention.
[00125] It should further be appreciated that although the invention has been described in the context of transform encoding of images, the invention is not limited to such contexts and may be utilized in various other compression applications and systems.
Furthermore, the invention is not limited to any one type of architecture or type of transform encoding.
For example, the DCT is mentioned above as one transform. Other transforms may be used, e.g., the new H.264/MEG-4 part 10 AVC video coding standard/draft standard defines 4 x 4 blocks and a DCT-like 4 x 4 integer transform. The invention does not depend on any particular type of interframe coding if used, or of motion compensation if used for interframe coding, or any intra-estimation if used for estimating the pixels of a block using information from neighboring blocks.
[00126] Note that variable length coding is sometimes referred to as entropy coding or statistical coding.
[00127] Note also that the term amplitude is irrespective of sign. Therefore, for example, coefficient of values +1 and -1 both have amplitude 1.

[00128] While one embodiment described herein includes a memory that stores a coding table, other embodiments store the coding information in the form of a data structure other than a table, e.g., a structure that includes a tree. Other data structures also may be used.
Similarly, while one embodiment described herein includes a memory that stores a decoding, other embodiments store the decoding information in the form of a data structure other than a table for the decoding.
[00129] Note also that the present invention does not depend on the particular type of VLC
used, and can work, for example, with Huffman coding and with arithmetic coding methods. Furthermore, while embodiments have been described that used fixed encoding for the events based on assumed or a-priori likelihoods of occurrence of the events (also called the symbols), i.e., the likelihoods of occurrence of the events do not change, other embodiments use adaptive encoding, i.e., the encoding is changeable according to statistical data such as histograms collected from the actual coefficients.
[00130] Thus, while there has been described what is believed to be the preferred embodiments of the invention, those skilled in the art will recognize that other and further modifications may be made thereto without departing from the spirit of the invention, and it is intended to claim all such changes and modifications as fall within the scope of the invention.

Claims (57)

1. A method for processing a series of digital signals that each have an amplitude from the amplitude set consisting of a first subset of values and a second subset of values, the first subset of values including the most common amplitude and at least one other amplitude, the processing to reduce the amount of data used to represent the digital signals and to form codewords such that the more frequently occurring sequences of values of digital signals are represented by relatively short codewords and the less frequently occurring sequences of values of digital signals are represented by relatively long codewords, the method comprising, forming a codeword for each of a set of modes encountered in the series of digital signals, a distinct mode defined for each amplitude of the second set of values; and for each encountered mode, forming a codeword for the events in the series of signals for the encountered mode, each event of each mode that is defined by an encountered amplitude of the second set of values being the runlength of none or more signals having the most common amplitude followed by such encountered amplitude when all amplitudes other than the most common amplitudes and the encountered amplitude are considered as if they had the most common encountered amplitude, such that relatively short codewords are formed to represent modes that are relatively more likely to occur, and within any mode, relatively short codewords are formed to represent events that are relatively more likely to occur.

2. A method as recited in claim 1, wherein the first subset of values consists of the most common amplitude and at least one other amplitude, and wherein a first mode is defined for the amplitude of the first subset other than the most common amplitude, such that each event of each mode is defined by an encountered amplitude other than the most common amplitude, and such that an event for each mode is defined by the runlength of a run of non-or more consecutive most common amplitudes that precede an encountered amplitude other than the most common when all amplitudes other than the most common amplitudes and the encountered amplitude are considered as if they had the most common encountered amplitude.

3. A method as recited in claim 1, wherein a first mode includes one or more sequences of a predefined set of sequences of one or more signals each having amplitude from the first subset of amplitudes, starting with an amplitude other than the most common amplitude, each event in the first mode being a run of none or more signals having the most common amplitude followed by a run of one or more identical sequences of the predefined set of sequences as if all amplitudes of the second set of values are equal to the most common amplitude.

4. A method as recited in claim 3, wherein a table is provided for storing a plurality of codewords for a set of events of the first mode, the table organized such that the more likely-to-occur events have shorter codewords than the less likely-to-occur events, and wherein the step of forming a codeword for each formed event of the first mode includes looking up the table.

5. A method as recited in claim 1, wherein the most common value is the amplitude most likely to occur, and wherein the other amplitudes in the first subset of amplitudes are more likely to occur than the amplitudes in the second subset of amplitudes.

6. A method as recited in claim 1, wherein a first mode includes all possible sequences having the length of the series of signals and having amplitude from the first subset of amplitudes, each event of the first mode being one of the sequences having amplitude from the first subset when all amplitudes of the second subset are considered equal to the most common amplitude.

7. A method as recited in claim 1, further comprising the step of:

forming fixed length codewords for at least some of the events for at least some of the modes.

8. A method as recited in claim 7, wherein the events for which the fixed codewords are formed are less likely to occur than the events for which a variable length codeword is used, and wherein the fixed length codewords are at least as long as any variable length codeword.

9. A method as recited in claim 1, wherein the most common amplitude is zero and the first subset of values includes amplitude 1.

10. A method as recited in claim 1, wherein the series of digital signals is a series of quantized coefficients of a transformed block of image data, the transform such that the most common amplitude is 0, wherein the first subset consists of the amplitudes 0 and 1, and wherein the second subset consists of the non-zero amplitudes that are greater than 1, such that the predefined set of sequences includes a single amplitude greater than one, or a sequence of amplitudes of 1's and 1's starting with a 1, and such that each event for which a codeword is formed is a run of none or more 0's followed by a run of either identical sequences of 0's and 1's that start with a 1 or a single amplitude greater than 1.

11. A method as recited in claim 5, wherein the forming of codewords for each event is fixed based on assumed a-priori likelihood of occurrence of the events.

12. A method as recited in claim 5, wherein the forming of codewords for each event is changeable according to actual statistical data collected from the series of digital values.

13. A method as recited in claim 1, further comprising:
forming an indication for when all the remaining digital signals in the series have the most common amplitude.

14. A method as recited in claim 1, wherein forming an event further includes indicating whether or not the event is the last for the series such that all the digital signals in the series following the event have the most common amplitude.

15. A method as recited in claim 1, further comprising the step of encoding said formed codewords for the events with information indicative of the sign of any non-zero amplitudes in the event.

16. A method of encoding an ordered quantized series of coefficients, the coefficients being of quantized coefficients of the transform of a block of image data, the quantized coefficients ordered along a two-dimensional path to form the series of quantized transform coefficients, the method comprising:
defining a set of one or more modes, including a distinct mode for and by each distinct encountered non-zero-amplitude in the series of coefficients;
forming a codeword to indicate each encountered mode;
forming events for each mode in a mode-modified series of coefficients, the mode-modified series formed by setting to zero all non-zero coefficients in the series of coefficients other than the coefficient amplitude defining the mode, each event in at least some of the modes that are defined by a non-zero-amplitude including a run in the mode-modified series of none or more zero-valued coefficients followed by the non-zero amplitude defining the mode;
forming a codeword for each formed event, forming a series of codewords that include the codewords to indicate the mode and for each mode the codewords to indicate the events for the mode;
such that relatively short codewords are formed to represent modes and events within modes that are relatively more likely to occur.

17. A method as defined in claim 16, wherein each mode is defined by a non-zero-amplitude in the series of coefficients and wherein events in each mode are defined by the runlengths of runs in the mode-modified series of none or more zero-valued coefficients followed by the non-zero amplitude defining the mode.

18. A method as recited in claim 16, wherein a first mode includes one or more sequences of a predefined set of sequences of one or more coefficients each having amplitude 0 or 1, starting with an amplitude of 1, and wherein an event in the first mode is a run in the mode-modified series of none or more zero-amplitude coefficients followed by a run of one or more identical sequences of the predefined set of sequences.

19. A method as recited in claim 18, wherein a table is provided for storing a plurality of codewords for a set of events of the first mode, the table organized such that the more likely-to-occur events have shorter codewords than the less likely-to-occur events, and wherein the step of foaming a codeword for each formed event of the first mode includes looking up the table.

20. A method as recited in claim 16, wherein a first mode includes all possible sequences having the length of the series of coefficients, having at least one coefficient with amplitude 1, and having all other coefficients with an amplitude of 0 or 1, and wherein an event of the first mode is the mode-modified series for the first mode.

21. A method as recited in claim 20, wherein a table is provided for storing a plurality of codewords for a set of events of the first mode, the table organized such that the more likely-to-occur events have shorter codewords than the less likely-to-occur events, and wherein the step of forming a codeword for each formed event of the first mode includes looking up the table.

22. A method as recited in claim 16, further comprising:
forming an indication for when all the remaining coefficients in the series have amplitude 0.

23. A method as recited in claim 16, wherein forming an event further includes indicating whether or not the event is the last for the series such that all the remaining coefficients in the series after the event have amplitude 0.

24. A method as recited in claim 16, further comprising the step of encoding said formed codewords of the event with an indication of the sign of any non-zero coefficients of the event.

25. A method as recited in claim 24, wherein the indication of the sign includes a sign bit for each codeword to indicate the sign of the non-zero coefficient.

26. A method as recited in claim 24, wherein a first mode defined by amplitude 1 includes one or more sequences of a predefined set of sequences of one or more coefficients each having amplitude 0 or 1, starting with an amplitude of 1, with a value of +1 distinguished from a value of -1, wherein a value of +1 is distinguished from a value of -1 in the mode modified series for amplitude 1, and wherein an event in the first mode is a run, in the mode-modified series, of none or more consecutive zero-amplitude coefficients followed by a run of one or more identical sequences of the predefined set of sequences of values 0, +1, and -1, starting with a value having amplitude 1.

27. A method as recited in claim 24, wherein a first mode defined by amplitude 1 includes any sequence in the set of all possible sequences of the same number of coefficients as in the series, each coefficient having amplitude 0 or 1, each sequence in the set including at least one non-zero amplitude, with a value of +1 distinguished from a value of -1, wherein a value of +1 is distinguished from a value of -1 in the mode modified series for amplitude 1, and wherein an event in the first mode is a run, in the mode-modified series, of none or more zero-amplitude coefficients followed by a run of one or more identical sequences of the predefined set of sequences of values 0, +1, and -1, starting with a value having amplitude 1.

28. A method as recited in claim 24, wherein the indication of the sign includes information related to a change of sign.

29. A method as recited in claim 24, wherein the information related to a change of sign is for any change of sign of non-zero-coefficients in the series of coefficients.

30. A method as recited in claim 24, wherein the information related to a change of sign is for any change of sign of non-zero-coefficients in the mode related series of coefficients.

31. A method as recited in claim 28, wherein a predefined initial sign is assumed.

32. A method as recited in claim 28, wherein an indication is included of sign of the first non-zero coefficient.

33. A method as recited in claim 24, wherein the indication of the sign includes an encoding of the sequence of signs of all the non-zero coefficients to produce a separate sign codeword for all the signs.

34. An apparatus to encode a series of quantized coefficients of a transform of a block of image data, the transform such that 0 is the most likely coefficient amplitude and 1 is the next most likely coefficient amplitude, the apparatus comprising:
means for recognizing any of a set of one or more defined modes, including a distinct mode for and defined by each distinct encountered non-zero-amplitude in the series of coefficients;
means for forming a codeword to indicate each encountered mode;
means for forming events for each mode in a mode-modified series of coefficients, the mode-modified series formed by setting to zero all non-zero coefficients in the series of coefficients other than the coefficient amplitude defining the mode, each event in at least some of the modes that are defined by a non-zero-amplitude including a run in the mode-modified series of none or more zero-valued coefficients followed by the non-zero amplitude defining the mode;
means for forming a codeword for each formed event, means for forming a series of codewords that include the codewords to indicate the mode and for each mode the codewords to indicate the events for the mode;
such that relatively short codewords are formed to represent modes and events within modes that are relatively more likely to occur.

35. An apparatus as defined in claim 34, wherein each mode is defined by a non-zero-amplitude in the series of coefficients such that events in each mode are defined by the runlengths of runs in the mode-modified series of none or more zero-valued coefficients followed by the non-zero amplitude defining the mode.

36. An apparatus as recited in claim 34, wherein a first mode includes one or more sequences of a predefined set of sequences of one or more coefficients each having amplitude 0 or 1, starting with an amplitude of 1, and wherein an event in the first mode is a run in the mode-modified series of none or more zero-amplitude coefficients followed by a run of one or more identical sequences of the predefined set of sequences.

37. An apparatus as recited in claim 34, wherein a first mode includes all possible sequences having the length of the series of coefficients, having at least one coefficient with amplitude 1, and having all other coefficients with an amplitude of 0 or 1, and wherein an event of the first mode is the mode-modified series for the first mode.

38. An apparatus as recited in claim 34, further comprising:
forming an indication for when all the remaining coefficients in the series have amplitude 0.

39. An apparatus as recited in claim 34, further comprising the step of encoding said formed codewords of the event with an indication of the sign of any non-zero coefficients of the event.

40. An apparatus as recited in claim 34, wherein the means for forming the codeword includes means for storing a table containing a plurality of codewords for a set of events of at least one of the modes, the table organized such that the more likely-to-occur events have shorter codewords than the less likely-to-occur events, and means for looking up the table.

41. An apparatus to encode a series of quantized coefficients comprising:
an event recognizer to recognize modes encountered in the series, each mode defined by a non-zero amplitude, the event recognizer also to recognize events in the mode-modified series formed from the series by setting to 0 all coefficients other than that defining the mode, an event for at least some of the modes being defined by the runlength of a run of none or more zero-valued coefficients, followed by a single coefficient of the amplitude defining the mode, an encoder coupled to the event recognizer, the encoder to form a codeword for each encountered mode and codewords for the recognized events in the encountered mode, such that the apparatus forms relatively short codewords to represent combinations that are relatively more likely to occur.

42. An apparatus as defined in claim 41, wherein an event in the mode defined by amplitude 1 is a run of none or more zero-valued coefficients, followed by a run of one or more of a predefined set of sequences, each sequence in the set of sequences being a sequence of 1's and 0's starting with a coefficient of amplitude 1, such that, for each event, the recognizer determines the runlength of the run of none or more zero-valued coefficients, in the case of the mode defined by amplitude 1, recognizer the sequence of the predefined set of sequences, and determines the runlength of the recognized sequence in the case that the run of none or more zero-valued coefficients is followed by a run of one or more of the sequences of the predefined set of sequences.

43. An apparatus as defined in claim 42, wherein the encoder includes a lookup device coupled to the event recognizer and determining the codeword for at least some of the events by looking up a table of codewords for combinations of runlengths of 0's, recognized sequences, and runlengths of the recognized sequence, the table stored in a memory.

44. An apparatus as defined in claim 41, wherein the event recognizer is such that each event of a first mode defined by amplitude 1 is a run in the mode-modified series of none or more zero-valued coefficients followed by one of a predefined set of sequences of 1's and 0's starting with a coefficient of amplitude 1.

45. An apparatus as recited in claim 41, wherein the encoder forms fixed length codewords for at least some of the events.

46. An apparatus as recited in claim 41, wherein the event recognizer further forms an indication for when all the remaining coefficients in the series have amplitude 0.

47. A method to decode a bitstream in order to form a series of quantized coefficients of a transform of a block of image data, the transform such that 0 is the most likely coefficient amplitude and 1 is the next most likely coefficient amplitude, the bitstream encoded by an encoding method applied to a series of coefficients that includes:
forming a codeword for each mode defined by a non-zero coefficient amplitude, an event for each mode being an event in the mode modified series formed by setting amplitudes other than that defining the mode to zero, an event in at least some of the modes being a run of none or more zero-valued coefficients followed a single coefficient of amplitude greater than one, such that relatively short codewords in the bitstream represent events that are relatively more likely to occur, the decoding method comprising:
recognizing a codeword for a mode;
recognizing a codeword for an event in the mode;
determining the event represented by the recognized codeword for the event; and determining the sub-series of coefficients of each determined event, until all coefficients in the series are determined.

48. A method as defined in claim 47, wherein an event in a first mode defined by amplitude 1 is a run of none or more zero-valued coefficients followed by a run of one of a predefined set of sequences, each sequence in the set being a sequence of 1's and 0's starting with a coefficient of amplitude 1.

49. A method as defined in claim 47, wherein an event in a first mode defined by amplitude 1 is a series of coefficients of amplitude 0 and 1 of the same length as the series of coefficients to which the encoding method was applied.

50. A method as defined in claim 48, wherein each event in the fist mode is a run of none or more zero-valued coefficients followed by one of the predefined set of sequences.

51. A method as recited in claim 48, wherein a table is provided containing a plurality of codewords for the first mode and the events for at least some of the codewords for the first mode, and wherein the determining of the event for a recognized codeword in the first mode includes looking up the table.

52. A method as recited in claim 47, wherein the coding is such that fixed length codewords are used to represent a subset of the events.

53. An apparatus to decode a bitstream encoded by an encoding method, the apparatus comprising:
a codeword recognizer to accepts the bits of the bitstream and recognize a codeword of a set of possible codewords that each represent a mode, the codeword recognizer further to recognize an event for each mode, a decoder coupled to the codeword recognizer to determine the event for the codeword recognized for the mode by the codeword recognizer, wherein the encoding method is applied to a series of coefficients and includes:

forming a codeword for each mode defined by a non-zero coefficient amplitude, an event for each mode being an event in the mode-modified series formed by setting amplitudes other than that defining the mode to zero, an event in at least some of the modes being a run of none or more zero-valued coefficients followed a single coefficient of amplitude greater than one, such that relatively short codewords in the bitstream represent events that are relatively more likely to occur.

54. An apparatus as defined in claim 53, wherein an event in a first mode defined by amplitude 1 is a run of none or more zero-valued coefficients followed by a run of one of a predefined set of sequences, each sequence in the set being a sequence of 1's and 0's starting with a coefficient of amplitude 1.

55. An apparatus as defined in claim 53, wherein an event in a first mode defined by amplitude 1 is a series of coefficients of amplitude 0 and 1 of the same length as the series of coefficients to which the encoding method was applied.

56. An apparatus as defined in claim 54, wherein each event in the fist mode is a run of none or more zero-valued coefficients followed by one of the predefined set of sequences.

57. An apparatus as recited in claim 54, wherein the decoder includes a lookup device that for the first mode looks up a table stored in a memory, the table providing a plurality of codewords for the first mode and the events for at least some of the codeword for the first mode.