CN1645750A - Coding and decoding method for variable long code - Google Patents

Abstract

The method includes following steps: the message source symbol needing to code are rearranged by the principle from big to small according to probability of occurrence of the message source symbol; the corresponding coding serial numbers are distributed; the code word layer and code word length of encoding is figured out by using eh coding serial numbers. the code word corresponding to coding serial numbers is generated. After receiving a code flow of inputted variable length code, the code word layer and DECINFO field are obtained from received variable length code; then the corresponding coding serial number is confirmed according to the coding word layer and DECINFO field; finally according to coding serial numbers the corresponding message source symbol is obtained.

Description

A kind of Code And Decode method of variable length code

Technical field

The present invention relates to the source encoding field, relate in particular to a kind of Code And Decode method of the variable length code in the extensive use of source encoding field.

Background technology

Variable-length encoding is a kind of destructive data compressing method, and it reduces the required mean code length of each symbol of coding by concentrate the big more short more code word of allocation of symbols length of probability of occurrence to source symbol, thereby improves overall compression efficiency.The probability match that code word size and corresponding source symbol occur must be good more, and code efficiency is just high more.Variable-length encoding is the important component part in modern image and the video coding system.In actual applications, not only require the compression efficiency height of variable length code, and require its encoding and decoding rule simple as far as possible.

Coefficient in image and the video coding after discrete cosine transform (DCT), quantification distributes and can regard the Generalized Gaussian information source as.In existing variable length encoding method, Golomb-Rice (GR) sign indicating number [Rice R F.Somepractical universal noiseless coding techniques.Tech.Rep.JPL.CA Pasadena:Jet Propulsion Laboratory, l979:79-22] and Exp-Golomb (EG) sign indicating number [Teuhola J.Acompression method for clustered bit-vectors.Inf.Process.Lett, 1978,10 (7): 308-311] be at present can be preferably therewith probability distribution be complementary the simple relatively again two kinds of code words of encoding and decoding rule.

The GR sign indicating number is to be the code word of parameter with suffix length (being designated as k), and its prefix length is a function of a single variable, and suffix has fixing length k.The code word size of GR sign indicating number is linear growth, and the number of codewords with same code word length is changeless.Therefore, the GR sign indicating number is a kind of quasi-optimal coding at geometric distributions, just at the optimum code of the Gaussian source (being Laplce's information source) of υ=1 (υ is the form parameter of Gaussian source).But along with reducing of υ value, generalized Gaussian distribution has more sharp-pointed peak value and slower hangover, so the efficient of GR sign indicating number can be more and more lower.

The EG sign indicating number is a kind of different prefix variable-length encoding.It is actually a kind of special shape [Elias P.Universal codeword sets and representations of the integers.IEEE Trans.on Inf.Theory, 1975,3 (21): 194-203] of the code word of Elias proposition.The number of the code word of the identical code length of EG sign indicating number is exponential increase along with the growth of code length.Therefore, the EG sign indicating number can characterize the generalized Gaussian distribution of 0＜υ＜1 with sharp-pointed peak value preferably, and reducing along with the υ value, the EG sign indicating number is more and more obvious with respect to the advantage of GR sign indicating number, but the EG sign indicating number approaches 1 quantification Generalized Gaussian information source for the υ value, and its optimum efficiency does not reach the performance of GR sign indicating number.

No matter this shows, be the EG sign indicating number, or the GR sign indicating number, all can not obtain the optimum code performance on the whole interval of υ value.

In order on the whole interval of υ value, to obtain the optimum code performance, need to propose a kind of new length variable code coding and decoding method.

Summary of the invention

An object of the present invention is to provide a kind of Code And Decode method of variable length code.This method can obtain good coding efficiency on the whole interval of υ value, and can carry out Code And Decode with simple algorithm.

The coding method of a kind of variable length code provided by the invention, its step comprises:

(A) establishing the source symbol collection that will encode is S={s ₀, s ₁..., s _N-1, corresponding probability of occurrence is distributed as P={P ₀, P ₁..., P _N-1, wherein N is the number of source symbol, according to probability principle from big to small the symbol among the source symbol collection S is resequenced, and distributes corresponding codes sequence number c to be respectively 0,1 ..., N-1;

(B) establish intermediate variable M, make M=c/2 ^k, wherein k is the code word size of suffix code, calculates the level j and the code word size l of MG sign indicating number according to the following rules:

(B1) if M equals 0, then make j=0, l=1+k changes step (C) over to;

(B2) if M equals 1, then make j=1, l=2+k changes step (C) over to;

(B3) if M greater than 1, establishes intermediate variable t and s, and makes j=1, t=1, s=1, carry out following steps again:

(B31) repeated assignment of values process: s=s * 2, t=t+s+1, j=j+1 changes step (B32) over to when t 〉=M;

(B32) if t 〉=(M+2), then l=2 * j+k changes step (C) over to; Otherwise l=2 * j+k+1 changes step (C) over to;

(C) as if j=0 or 1, no information field INFO, the suffix of k bit are 0～(2 ^k-1) binary representation; If j＞1, the decimal value of computing information field DECINFO is designated as [DECINFO] according to the following rules ₁₀:

If l=2 * j+k, then [DECINFO] ₁₀=c-(t-s) * 2 ^k

If l=2 * j+k+1, then [DECINFO] ₁₀=c-(t+1-2 * s) * 2 ^k

(D) the resulting MG code word corresponding with the coding sequence number is:

[j 0] [1] [INFO] [suffix of k bit], wherein, the part that is made of the suffix of INFO field and k bit is designated as DECINFO.

The coding/decoding method of above-mentioned coding, its step comprises:

(A), receive an encoding code stream at receiving terminal;

(B) read 0 from the encoding code stream that obtains, up to running into till 1, it is j that note is read 0 number, and proceeds according to the following rules:

If j≤1 is then read k bit again, wherein k is the code word size of suffix code, and this k bit is designated as DECINFO;

If j＞1 is then read j+k-1 bit again, and the decimal value corresponding with it is designated as MEDINFO; If MEDINFO 〉=(2 ^K+j-1-2 ^k), then read a bit again, this j+k bit is designated as DECINFO, its code length l=2 * j+k+1; If MEDINFO＜(2 ^K+j-1-2 ^k), then this j+k-1 bit being designated as DECINFO, its code length is l=2 * j+k;

(C) decimal value of note DECINFO field is [DECINFO] ₁₀, calculate coding sequence number c according to the following rules:

If j=0, then c=[DECINFO] ₁₀

If j=1, then c=2 ^k+ [DECINFO] ₁₀

If j＞1, when l=2 * j+k, then $c=\left[\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-2}+1)\right]\×2^k+{\left[\mathrm{DECINFO}\right]}_{10};$ Otherwise,

$c=[2\×j-2^i+\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-1}-1)]\×2^k+{\left[\mathrm{DECINFO}\right]}_{10};$

Calculate the coding sequence number, obtain corresponding source symbol according to the coding sequence number again.

The present invention combines the advantage of EG and GR sign indicating number, and each layer code word all comprises the code word of two kinds of code lengths, and wherein a kind of number of codewords is exponential increase with the increase of the number of plies, and another kind then remains unchanged; Thereby the quantification Generalized Gaussian information source for form parameter changes in a wider context also has stable coding efficiency.And the variable length code that the present invention obtained has fairly regular structure, can realize encoding and decoding by simple algorithm, thereby avoids transmitting and the storage code table, has saved transmission bandwidth and memory space.

Description of drawings

Fig. 1 shows the schematic flow sheet according to the coding method of variable length code of the present invention;

Fig. 2 shows the schematic flow sheet according to the coding/decoding method of variable length code of the present invention.

Embodiment

The present invention at first according to the probability distribution of source symbol, resequences according to the symbol that probability of occurrence principle is from big to small concentrated the source symbol that will encode, distributes the corresponding codes sequence number; The coding sequence number that utilization obtains calculates the code word level and the code word size of coding; After obtaining code word level and code word size, produce and the corresponding code word of this coding sequence number.And after receiving an input variable length sign indicating number code stream; From the variable length code code stream that is received, draw code word level and information field DECINFO; Utilize the code word level and the information field DECINFO that are drawn to determine the corresponding codes sequence number then; Obtain corresponding source symbol according to the coding sequence number at last.

Before the present invention is described in further detail, at first the formation of variable length code (hereinafter to be referred as the MG sign indicating number) is simply introduced.

A code word in the MG sign indicating number is made of prefix code and suffix code two parts usually, and the code word size of suffix code is the k position.K is a prior given nonnegative integer, is used for controlling the number of every layer of code word.K is unsuitable excessive, and the present invention arranges k＜256.The MG sign indicating number is made up of several layers (number of plies is designated as j), and the number of every layer of code word is provided with as follows:

1 for the shortest two-layer code word of MG sign indicating number, i.e. j=0 and j=1, and number of codewords remains 2 ^kConstant.

2 in j＞1 o'clock, and there is the code word of two kinds of code lengths in each layer of MG sign indicating number, and promptly code length is 2j+k or 2j+k+1.For code length is the code word of 2j+k, and its number of codewords is exponential increase with number of plies increase, and number is 2 ^K+j-1-2 ^kAnd be the code word of 2j+1+k for code length, keep number of codewords constant, be 2 ^K+1

When the suffix length of MG sign indicating number is made as k, 2 of ground floor (when being j=0) ^kThe prefix of individual code word is " 1 ", 2 of the second layer (when being j=1) ^kThe prefix of individual code word is " 01 ".The code word of the ground floor and the second layer all be separately prefix and the suffix of k bit be formed by connecting, wherein the suffix of k bit is 0～(2 ^k-1) binary representation.For the code word of all the other levels (be j＞1 o'clock), produce by following rule:

(1) for code length be the code word of 2j+k, promptly prefix is the code word of even bit bit, is connected with one 1 with j 0 earlier, and then is connected with j+k-1 bit and constitutes.Wherein, a back j+k-1 bit is 0～(2 ^K+j-1-2 ^k-1) binary representation.

(2) for code length be the code word of 2j+1+k, promptly prefix is the code word of odd bits bit, links to each other with one 1 with j 0 earlier, and then is connected with j+k bit and constitutes.Wherein, a back j+k bit is (2 ^K+j-2 ^K+1)～(2 ^K+j-1) binary representation.

In a word, being constructed as follows of MG code word:

[j individual 0] [1] [INFO] [suffix of k bit]

Wherein, ground floor code word (being j=0) does not have preposition 0 and the INFO field.Second layer code word (being j=1) have one preposition 0, but do not have information field INFO.When j＞1, if code word size is 2j+k, then information field INFO is a j-1 bit, and its value is 0～(2 ^J-1-2) binary representation; If code word size is 2j+1+k, then information field INFO is a j bit, and its value is for (2 ^j-2)～(2 ^j-1) binary representation.The suffix of k bit then is 0～(2 ^k-1) binary representation.For simplicity, the part that is made of the suffix of INFO field and k bit in the MG code word is designated as DECINFO.

Specify the Code And Decode method of variable length code below in conjunction with attached Fig. 1 and 2.

The coding method of MG sign indicating number

As shown in Figure 1, suppose that the source symbol collection that will encode is S={s ₀, s ₁..., s _N-1, corresponding probability of occurrence is distributed as P={p ₀, p ₁..., p _N-1, wherein N is the number of source symbol.According to probability principle from big to small the symbol among the source symbol collection S is resequenced, distributes corresponding codes sequence number (representing) to be respectively 0,1 with c ..., N-1 (step S10).Wherein, the symbol corresponding codes sequence number of probability maximum is 0, and second largest symbol corresponding codes sequence number of probability is 1, and the rest may be inferred, and the symbol corresponding codes sequence number of probability minimum is N-1.

The structure of MG sign indicating number is relevant with the suffix length k of the hierachy number j at code word place and code word.Obtain the code word of MG sign indicating number, at first will calculate j by coding sequence number c.In addition, because the code word of same level has two classes: length is respectively 2j+k and 2j+1+k.Therefore, also need the affiliated classification of definite code word.The specific coding step is as follows:

The first step is calculated the level j and the code word size l (step S20) of corresponding MG sign indicating number according to coding sequence number c.Make M=c/2 ^k, calculate according to the following rules:

(1) if M equals 0, then make j=0, l=1+k changed for second step over to.

(2) if M equals 1, then make j=1, l=2+k changed for second step over to.

(3) if M greater than 1, then j and l are obtained by following step: (establish intermediate variable t and s, wherein t be used for the expression add up and.In fact, the value of t is the coding sequence number of last code word of j layer).

When (3.1) beginning, make j=1, t=1, s=1;

(3.2) repeated assignment of values process: s=s * 2, t=t+s+1, j=j+1 changes step over to when t 〉=M

(3.3) proceed;

(3.3) if t 〉=(M+2), then l=2 * j+k; Otherwise, l=2 * j+k+1;

In second step, after obtaining j and l, produce corresponding code word (step S30) with coding sequence number c according to following method.Before j+1 bit be that j individual 0 links to each other with one 1 and forms, a back j-1+k/j+k bit is the suffix of information field INFO and k bit, i.e. information field DECINFO, wherein, the decimal value of information field DECINFO is designated as [DECINFO] ₁₀Its computation rule is as follows:

If j=0 or 1, no information field INFO, the suffix of k bit are 0～(2 ^k-1) binary representation;

If j＞1, then [DECINFO] ₁₀Obtain by following steps:

If l=2 * j+k, then [DECINFO] ₁₀=c-(t-s) * 2 ^k

If l=2 * j+k+1, then [DECINFO] ₁₀=c-(t+1-2 * s) * 2 ^k

Thus, just obtaining the MG code word corresponding with the sequence number of encoding is:

[j 0] [1] [INFO] [suffix of k bit], wherein, the part that is made of the suffix of INFO field and k bit is DECINFO.

The coding/decoding method of MG sign indicating number

As shown in Figure 2, at receiving terminal, decoder is decoded to input code flow, draws the coding sequence number.

Decoding step is as follows:

At first, at receiving terminal, receive an encoding code stream (step S40).

Secondly, based on the input code flow that receives, obtain the level j and the DECINFO field (step S50) of MG sign indicating number.Promptly read 0 from the encoding code stream that obtains, up to running into till 1, it is j that note is read 0 number, and proceeds according to the following rules:

(1) if j≤1 is then read k bit again, this k bit is designated as DECINFO.

And the decimal value corresponding with it is designated as MEDINFO (2) otherwise read j+k-1 bit again.If MEDINFO 〉=(2 ^K+j-1-2 ^k), then read a bit again.This j+k bit is designated as DECINFO.At this moment, code length is l=2 * j+k+1.If MEDINFO＜(2 ^K+j-1-2 ^k), then this j+k-1 bit being designated as DECINFO, the code length of code word is l=2 * j+k.

Then, (decimal value of DECINFO field is designated as [DECINFO] obtaining j and DECINFO ₁₀) after, calculate coding sequence number c (step S60) according to the following rules, that is:

If j=0, then c=[DECINFO] ₁₀

If j=1, then c=2 ^k+ [DECINFO] ₁₀

If j＞1 can calculate c according to following steps:

If l=2 * j+k, then $c=\left[\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-2}+1)\right]\×2^k+{\left[\mathrm{DECINFO}\right]}_{10};$ Otherwise,

$c=[2\×j-2^i+\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-1}-1)]\×2^k+{\left[\mathrm{DECINFO}\right]}_{10}.$

Thus, by calculation expression, just can obtain the corresponding codes sequence number according to encoding code stream.Obtain corresponding source symbol according to the coding sequence number again, so far, decode procedure is finished.

Example

Table 1 has provided an object lesson (k=0) of MG sign indicating number.

By the content of front as can be known, code length is that the scope of the code word corresponding codes sequence number of 2j+k (j＞1) is:

$[2+2\×(j-2)+\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-2}-1)]\×2^k~[2+2\×(j-2)+\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-1}-1)]\×2^k-1.$

And code length is the scope of the code word corresponding codes sequence number of 2j+k+1 (j＞1) be:

$[2+2\×(j-2)+\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-1}-1)]\×2^k~[4+2\×(j-2)+\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-1}-1)]\×2^k.$

Because the code word number of same code word length all is 2 ^kIntegral multiple, the sequence number of will encoding is divided by 2 ^k, the value of note gained is M, the length of the code word of the coding sequence number correspondence that then M is identical is identical, and the difference of these code words only is the suffix of k bit.That is to say that [j individual 0] [1] [INFO] part of the MG code word corresponding with original encoding sequence number c is come out by the MG code construction of coding sequence number M being carried out k=0.Above analysis result has provided the level characteristics of MG sign indicating number, and it is as shown in table 2.Wherein, k is the suffix length of MG sign indicating number; M=c/2 ^k(c is the coding sequence number).

An example (k=0) of table 1 MG sign indicating number

Coding sequence number (c)	Code word	Code length
			?????0	?1	?1
?????1	?01	?2
			?????2	?0010	?4
?????3	?00110	?5
			?????4	?00111	?5
?????5	?000100	?6
			?????6	?000101	?6
?????7	?000110	?6
			?????8	?0001110	?7
?????9	?0001111	?7
			?????10	?00001000	?8
?????……	?……	?……

The level characteristics of table 2 MG sign indicating number

Claims (2)

1, a kind of coding method of variable length code, its step comprises:

(A) establishing the source symbol collection that will encode is S={s ₀, s ₁..., s _N-1, corresponding probability of occurrence is distributed as P={p ₀, p ₁..., p _N-1, wherein N is the number of source symbol, according to probability principle from big to small the symbol among the source symbol collection S is resequenced, and distributes corresponding codes sequence number c to be respectively 0,1 ..., N-1;

(B) establish intermediate variable M, make M=c/2 ^k, wherein k is the code word size of suffix code, calculates the level j and the code word size l of MG sign indicating number according to the following rules:

(B1) if M equals 0, then make j=0, l=1+k changes step (C) over to;

(B2) if M equals 1, then make j=1, l=2+k changes step (C) over to;

(B3) if M greater than 1, establishes intermediate variable t and s, and makes j=1; T=1; S=1, carry out following steps again:

(B31) repeated assignment of values process: s=s * 2, t=t+s+1, j=j+1 changes step (B32) over to when t 〉=M;

(B32) if t 〉=(M+2), then l=2 * j+k; Otherwise, l=2 * j+k+1;

(C) as if j=0 or 1, no information field INFO, the suffix of k bit are 0～(2 ^k-1) binary representation; If j＞1, the decimal value of computing information field DECINFO is designated as [DECINFO] according to the following rules ₁₀:

If l=2 * j+k, then [DECINFO] ₁₀=c-(t-s) * 2 ^k

If l=2 * j+k+1, then [DECINFO] ₁₀=c-(t+1-2 * s) * 2 ^k

(D) the resulting MG code word corresponding with the coding sequence number is:

[j 0] [1] [INFO] [suffix of k bit], wherein, the part that is made of the suffix of INFO field and k bit is DECINFO.

2, a kind of coding/decoding method of variable length code, its step comprises:

(A), receive an encoding code stream at receiving terminal;

(B) read 0 from the encoding code stream that obtains, up to running into till 1, it is j that note is read 0 number, and proceeds according to the following rules:

If j≤1 is then read k bit again, wherein k is the code word size of suffix code, and this k bit is designated as DECINFO;

If j＞1 is then read j+k-1 bit again, and the decimal value corresponding with it is designated as MEDINFO; If MEDINFO 〉=(2 ^K+j-1-2 ^k), then read a bit again, this j+k bit is designated as DECINFO, its code length l=2 * j+k+1; If MEDINFO＜(2 ^K+j-1-2 ^k), then this j+k-1 bit being designated as DECINFO, its code length is l=2 * j+k;

(C) decimal value of note DECINFO field is [DECINFO] ₁₀, calculate coding sequence number c according to the following rules:

If j=0, then c=[DECINFO] ₁₀

If j=1, then c=2 ^k+ [DECINFO] ₁₀

If j＞1, when l=2 * j+k, then $c=\left[\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-2}+1)\right]\×2^k+[\mathrm{DECINFO}{]}_{10};$ Otherwise, $c=[2\×j-2^i+\underset{i=2}{\overset{i=j}{\mathrm{\Σ}}}(2^{i-1}-1)]\×2^k+{\left[\mathrm{DECINFO}\right]}_{10};$ Calculate the coding sequence number, obtain corresponding source symbol according to the coding sequence number again.

Images