CN101370138B - A Decoding Method of H.264 Standard CAVLC Residual Coefficients - Google Patents

Abstract

The invention discloses a decoding method for H.264 context adaptive coding residual coefficients, comprising the following steps: firstly, according to a designed three-dimensional code table, decoding the binary group of the number of non-zero coefficients and the number of trailing coefficients; Then decode the sign bit of the trailing coefficient; then decode the magnitude of the non-tailing non-zero coefficient; finally decode the number of zeros before the last non-zero coefficient and the number of zeros before each non-zero coefficient. Among them, the number of non-zero coefficients and the number of trailing coefficients are decoded. According to the three-dimensional code table, by prefetching a specific number of bits, the length of the codeword information bit can be determined by at most one comparison, and then the content of the two-tuple can be determined. . Among them, the binary group decoding of the number of non-zero coefficients and the number of tailing coefficients and the amplitude decoding of non-tailing non-zero coefficients adopt the method of reading two bytes at a time to obtain the number of leading 0s. The method can effectively improve the decoding speed of the CAVLC code stream.

Description

A decoding method of H.264 standard CAVLC residual coefficient

technical field

The invention relates to the field of video and image coding and decoding, in particular to a decoding method of H.264 context adaptive variable length coding (CAVLC) residual coefficients. the

Background technique

The latest international video coding standard H.264 stipulates that its entropy coding can adopt CAVLC or context-adaptive arithmetic coding (CABAC). The computational complexity is much lower than that of CABAC, and it has achieved a good compromise between coding efficiency and complexity. The H.264 protocol standard especially recommends this technology for basic layer applications such as video surveillance and video conferencing. These applications require real-time decoding of code streams, and in CAVLC, different decoding methods are used for different syntax elements: direct or indirect exponential Golomb codes are used for decoding macroblock types, motion vectors, etc., and the decoding of residual coefficients is due to the need Analyzing all the codewords generated in the encoding one by one is more complicated and the speed is lower. The codeword specifically includes the number of non-zero coefficients and the number of trailing coefficients, the binary group coeff_token<TotalCoeff, TrailingOnes>, the sign bit of the trailing coefficient, the magnitude of the non-zero coefficient of the non-tailing coefficient, and 0 before the last non-zero coefficient and the number of leading zeros for each nonzero coefficient. the

The residual coefficient decoding time is mainly consumed in the decoding of the binary group of the number of non-zero coefficients and the number of trailing coefficients. The decoding of the codeword not only needs to switch the code table, but also is mainly provided by the H.264 protocol standard. Due to the limitation of the one-dimensional code table, in the code word parsing process, it is necessary to read the code word bit by bit. Every time a bit is read, it must be traversed in the code table to determine whether the part that has been read is a valid code word, until Until the read-in bit string can match a certain code word in the code table, the reading operation and comparison operation need to be repeated in the decoding process, and the efficiency is very low. the

For the amplitude decoding of non-tailing non-zero coefficients, the determination of the number of leading 0s also consumes a lot of time, because it can only be judged bit by bit whether it is 0, until 1 is read. the

In view of the above problems, since the CAVLC algorithm was adopted in 2003, some implementation schemes have appeared. Document 1 (Hsiu-Cheng Chang, Chien-Chang Lin and Jiun-In Guo, "A Novel Low-Cost High-Performance VLSI Architecture for MPEG-4AVC/H.264 CAVLC Decoding", Proc.ISCAS 2005, pp.6110- 6113, 23-26 May 2005) disclosed the VLSI (Very Large Scale Integration) design of CAVLC decoding, document 2 (Xing Qin, Xiaolang Yan, "A memory and Speed Efficient CAVLC Decoder", Proc.VCIP 2005, pp .1418-1426, Jul.2005) discloses a programmable VLSI design, document 3 (Patent "A Method Based on Context-Adaptive Variable Length Decoding" applied by Tsinghua University, Publication No. 200610041780.X) based on this A design method of IC-oriented CAVLC decoding with lower storage requirements and faster speed is proposed. The above methods are all proposed for the integrated circuit design of CAVLC, rely on the integrated circuit design platform, and are not suitable for software implementation on platforms such as general-purpose CPUs or DSPs. Document 4 (Patent "H.264 Decoding Method for Fast Context-Adaptive Variable Length Coding", Publication No. 200610002800.2, filed by Korea C&S Technology Co., Ltd.) discloses a three-dimensional code table, which can be used for CAVLC software decoding. However, this method has a relatively large memory overhead and a large amount of calculation, so the calculation speed is relatively slow, and there is a possibility of further optimization. the

At the same time, the above documents are optimized for the early version of the code table. In the latest H.264 standard, the original code table is adjusted, and a code table with nC==-2 is added in particular. The invention designs a CAVLC decoding method aiming at the code table formulated by the new H.264 protocol standard. the

Contents of the invention

The object of the present invention is to improve the problems of large amount of computation and slow computation speed in the existing decoding method, and to provide a high-efficiency programmable decoding method of context-adaptive coding coefficients. the

In order to achieve the above object, according to one aspect of the present invention, a decoding method of the H.264 standard CAVLC residual coefficient is provided. Analyze one by one, including the following steps:

The number of non-zero coefficients and the number of trailing coefficients are decoded based on the code table, wherein the code table includes the number of leading 0s, the length of the information bit and the three-dimensional parameters of the information, and the length of different information bits corresponds to the reading according to the length of the information bit retrieved information;

Decode the sign bit of the trailing coefficient;

Decode the magnitude of non-smearing non-zero coefficients;

Decode the number of zeros before the last non-zero coefficient;

Decodes the number of leading zeros for each nonzero coefficient. the

According to another aspect of the present invention, the binary group decoding based on the number of the non-zero coefficients of the code table and the number of trailing coefficients further comprises the following steps:

Calculate nC according to the H.264 standard, and select the code table according to nC;

Read the number of leading 0s;

According to the selected code table and the number of leading 0s, the information bits are read to determine the binary group information of the number of non-zero coefficients and the number of trailing coefficients. the

According to another aspect of the present invention, reading the number of leading 0s further comprises the following steps:

Read 2 bytecode streams;

Read the highest number of digits whose value is 1 in the two bytecode streams, then the previous digits are the number of leading 0s. the

The present invention has the advantages of effectively reducing the number of operations for reading code streams; effectively reducing the code table storage space; effectively reducing the number of codeword matching times, and performing at most one match for the number of non-zero coefficients and the number of trailing coefficients. In most cases, the codeword can be determined without matching, which improves the decoding speed of the code stream. the

Description of drawings

Below, describe embodiment of the present invention in detail in conjunction with accompanying drawing, wherein:

Figure 1 shows the reverse zigzag scan used in residual coefficient coding. the

Figure 2 The positional relationship between nC and nA and nB. the

Figure 3 The code table used for encoding in the H.264 protocol standard. the

Figure 4 shows the codeword construction method of the Exponential Golomb code. the

FIG. 5 is a code table for decoding the number of non-zero coefficients and the number of trailing coefficients (0=<nC<2). the

FIG. 6 is a code table for decoding the number of non-zero coefficients and the number of trailing coefficients (2=<nC<4). the

FIG. 7 is a code table for decoding the number of non-zero coefficients and the number of trailing coefficients (4=<nC<8). the

FIG. 8 is a code table for decoding the number of non-zero coefficients and the number of trailing coefficients (8=<nC). the

Fig. 9 is a code table for the decoding of the number of non-zero coefficients and the number of trailing coefficients (nC==-1) tuples. the

Fig. 10 is a code table for the decoding of the number of non-zero coefficients and the number of trailing coefficients 2-tuple (nC==-2). the

Fig. 11 is a flow chart of decoding a binary group of the number of non-zero coefficients and the number of trailing coefficients. the

Fig. 12 is another non-zero coefficient number and the number binary group decoding flow chart of the trailing coefficient. the

Detailed ways

The decoding process of the residual coefficient of adaptive variable length coding based on the context of the H.264 protocol standard is a complex process, and the residual coefficient is scanned in reverse zigzag from high frequency to low frequency (as shown in Figure 1) After that, all codewords are analyzed one by one. Existing decoding methods usually include the following steps:

Step 1: Decode the binary group of the number of non-zero coefficients and the number of trailing coefficients, where the number of trailing coefficients specifically refers to the non-zero, absolute value in the one-dimensional coefficient matrix obtained by scanning from low frequency to high frequency The number of consecutive coefficients of 1. The maximum number of trailing coefficients is 3, and more than 3 ±1 coefficients are regarded as ordinary coefficients, and the decoding method adopts the decoding method of the number of 0s before each non-zero coefficient. the

The specific decoding steps include:

(1) Select the code table according to the context: the number of non-zero coefficients and the number of trailing coefficients The selection of the code table used for binary decoding needs to be based on the distribution of the residual coefficients of the surrounding blocks and depends on the value of nC. The specific calculation steps of nC are as follows: if the chroma DC coefficient is decoded, nC=-1 in 4:2:0 format, nC=-2 in 4:2:2 format, nC=0 in 4:4:4 format ; Otherwise, nC calculates according to the coefficient number nA of the upper side block and the coefficient number nB of the left block (nA, nB, nC position relation as shown in Figure 2): if nA and nB are all available, then nC=(nA+nB +1)/2; otherwise, nC=nA if nA is available, nC=nB if nB is available, otherwise nC=0. The number of non-zero coefficients and the number of trailing coefficients are as shown in Figure 3 for decoding the H.264 standard code table;

(2) Read the code stream bit by bit until the read-in bit string can match a certain code word in the code table; take out the number of non-zero coefficients and the number of tailing coefficients corresponding to this code word. the

Step 2: Decode the sign bit of the trailing coefficient, decode 1 bit for each sign, and decode it as 0 or 1 according to the sign. the

Step 3: Use the mapped Exponential Columbus code to decode the magnitude of the non-tailing non-zero coefficients according to the scanning order. This is a structured decoding method. The codeword construction method is shown in Figure 4. The specific steps include:

(1) Read the number of leading 0s: read bit by bit, if it is 0, continue to read the next bit; if it is 1, proceed to the next step;

(2) Calculate the magnitude of the non-zero coefficient of non-tailing

Wherein the information part is the part representing the amplitude in the code word in the code table (as shown in Figure 4), its length is equal to the leading 0 digits, and Ceil represents an upward rounding function. the

Step 4: Use an adaptive code table to decode the number of zeros before the last non-zero coefficient;

Step 5: adaptively select a code table according to the coded information, and decode the number of 0s before each non-zero coefficient. the

It can be seen from the above decoding process that due to the number of non-zero coefficients and the number of trailing coefficients, the binary decoding process involves a lot of bit-by-bit reading and codeword comparison operations, which will lead to a serious decrease in decoding speed. The defects in the decoding process are essentially determined by the one-dimensional characteristic of the H.264 standard code table. For this reason, the present invention has designed a kind of new code table, utilizes leading 0 digits, information bit length and information part three-dimensional parameter to design three-dimensional code table, can directly determine the number of non-zero coefficient and the number of trailing coefficient thus Two-tuple. the

Specifically, the present invention designs a three-dimensional code table based on codeword features for decoding the binary group of the number of non-zero coefficients and the number of trailing coefficients. By observing the H.264 standard code table in Figure 3, it is divided into 0=<nC<2, 2=<nC<4, 4=<nC<8, nC>=8, nC==-1 and nC= =-2 There are 6 code tables in total. For these six code tables, reorganize them respectively to construct six three-dimensional tables as shown in Fig. 5, Fig. 6, Fig. 7, Fig. 8, Fig. 9 and Fig. 10. A blank portion in the code table indicates that there is no corresponding code word. In the code table of the present invention, after the context and the number of leading 0s are determined, there are at most two lengths of information bits, and the difference between the length values is at most 1, which is very effective for the storage space of compressed codewords, and the storage space can be effectively compressed by using this point . In the decoding process, if there are two types of information bit lengths corresponding to the number of leading 0s, first consider the case of a large length, read in the number of corresponding bits, and determine the information bit length when it can match the code table ; When it cannot be matched with the code table, it is considered that the information bit has a small length. Especially, when nC＞=8 (as shown in Figure 8), the code table is fixed-length, and codewords can be obtained without multiple comparisons. The present invention directly sets the length of its leading 0 to 0, and the following 6 bits All are treated as information bits. the

In the code table, the information can be the index index of the binary group of the number of non-zero coefficients and the number of trailing coefficients, or directly the specific content of the binary group of the number of non-zero coefficients and the number of trailing coefficients. The former can reduce the code The space occupied by the table, but the number of non-zero coefficients and the number of trailing coefficients need to be calculated according to the index, and the latter can directly obtain the required element value. You can choose one of them according to the resource constraints of different platforms. If the index is stored in the code table, the number of non-zero coefficients and the number of trailing coefficients are calculated as follows:

if (index<6) {

If (index!=0) then index+=3;

If (index>=3) then index+=2;

TotalCoeff＝(index＞＞2)&3;

TrailingOnes=index&3;

}

else {

TotalCoeff＝3+(index-6)/4;

TrailingOnes=(index-6)&3;

}

In order to reduce the time required to obtain the number of leading 0s during the decoding process of the number of non-zero coefficients and the number of trailing coefficients and the magnitude of non-tailing non-zero coefficients, the present invention designs a method for quickly obtaining leading 0s. number of programs. Considering that the leading 0 can be as long as 12 bits, and at the same time reduce the number of reading code streams as much as possible, when reading the leading 0 digits, read 2 bytes each time. In addition, many specific platforms have special instructions to implement this function, and making full use of these instructions can further reduce the cycle required to read the leading 0 digits. the

On the basis of the above-mentioned three-dimensional code table and the scheme for quickly obtaining the number of leading 0s, the present invention proposes a decoding method of the H.264 standard CAVLC residual coefficient, and the specific steps are as follows:

1. Two-tuple decoding of the number of non-zero coefficients and the number of trailing coefficients:

Assume that the current code stream points to "00000000010110110...". the

First: Select the code table according to the context

Calculate nC according to the H.264 standard, and determine the code table that should be selected according to nC. Assuming nC=1, the code table corresponding to the binary group decoding of the number of non-zero coefficients and the number of trailing coefficients should select the code table corresponding to 0=<nC<2 (FIG. 5). the

Second, read the number of leading 0s

Read the number of leading 0s from the current position of the code stream. The method is as follows: read the code stream of 2 bytes, and then read the highest digit whose value is 1, and the previous digit is the number of leading 0s. In this example, "0000000001011011" is read in, and the number of leading 0s is 9. the

Third, read the information bits to determine the number of non-zero coefficients and the number of trailing coefficients. Two-tuple information

According to the selected code table and the number of leading 0s, determine the possible information bit length, and read it through the shift method. In this example, it can be determined that the information bit length is 4, and 4 bits "1011" are read in by shifting. Binary "1011" corresponds to decimal 11. If the number of non-zero coefficients and the number binary group content of trailing coefficients are directly stored in the code table, then directly obtaining the number of trailing coefficients is 0, and the number of non-zero coefficients is 7 (as shown in Figure 11); if the table Stored in is the index of the codeword, then 22 can be obtained, and the number of trailing coefficients obtained by calculation is 0, and the number of non-zero coefficients is 7 (as shown in FIG. 12 ). the

2. The sign bit of the trailing coefficient is decoded, and it is decoded as positive or negative according to the code word 0 or 1;

3. Amplitude decoding of non-smearing non-zero coefficients

Assume that the current code stream points to "000101101101100101111001...". the

The first step is to read the number of leading 0s

Read the number of leading 0s from the current position of the code stream. The method is as follows: read the code stream of 2 bytes, and then read the highest digit whose value is 1, and the previous digit is the number of leading 0s. In this example, "0001011011011001" is read in, and the number of leading 0s is 3. the

The second step is to calculate the magnitude of the non-zero coefficient of non-tailing

4. Decode the first non-zero coefficient of the reverse scan;

5. Decode the number of zeros before each non-zero coefficient. the

Claims (5)

1. the coding/decoding method of standard context adaptive variable length coding (CAVLC) residual error coefficient H.264 comprises the following steps:

Residual error coefficient is carried out by the zigzag scanning of high frequency to low frequency;

Based on the number of code table decoding nonzero coefficient and two tuples of hangover number of coefficients, wherein, described code table comprises standard code table is H.264 cut apart six code tables that obtain according to 0=＜nC＜2,2=＜nC＜4,4=＜nC＜8, nC＞=8, nC==-1 and nC==-2; Described code table comprises three-dimensional parameter: leading 0 number, information bit length and information, described code table is the also corresponding information bit information that reads according to information bit length under different information bit length, when information bit length is identical, can correctly find the number and the hangover number of coefficients information of nonzero coefficient; This step further comprises:

A, calculate nC, select described code table according to nC according to standard code H.264;

B, read leading 0 number;

C, according to code table of selecting and leading 0 number, read information bit, determine the number of nonzero coefficient and the binary group information of hangover number of coefficients;

The sign bit of decoding hangover coefficient;

The decode amplitude of nonzero coefficient of non-hangover;

The decode number of last nonzero coefficient preceding 0;

The decode number of each nonzero coefficient preceding 0.

2. the method for claim 1 is characterized in that described information is the number of nonzero coefficient and two tuples of hangover number of coefficients.

3. the method for claim 1 is characterized in that described information is the index value of two tuples of the number of nonzero coefficient and the number of coefficients of trailing, and can calculate described two tuples according to this value, and concrete computational process is as follows:

(if index＜6) {

=0) index+=3 so;

If (index＞=3) are index+=2 so;

TotalCoeff＝(index＞＞2)&3；

TrailingOnes＝index&3；

}

Otherwise

TotalCoeff＝3+(index-6)/4；

TrailingOnes＝(index-6)&3；

}

Wherein, index is described index value, and TotalCoeff is that the number and the TrailingOnes of described nonzero coefficient is described hangover number of coefficients.

4. the method for claim 1 is characterized in that the amplitude of the nonzero coefficient of the non-hangover of described decoding further comprises the following step:

Read leading 0 number;

Calculate the amplitude of the nonzero coefficient of non-hangover according to leading 0 number that reads.

5. method as claimed in claim 4 is characterized in that described leading 0 the number of reading further comprises following steps:

Read 2 byte code streams;

Read value is 1 highest order number in these 2 byte code streams, and then Zhi Qian figure place is leading 0 number.

Images