CN102368385A - Backward block adaptive Golomb-Rice coding and decoding method and apparatus thereof - Google Patents

Abstract

The invention provides a backward block adaptive Golomb-Rice coding and decoding method. The method comprises the following steps: realizing backward adaptive Golomb-Rice coding by utilizing an adaptive update rule; updating an adaptive parameter by utilizing a sliding window method; carrying out block updating on a Golomb-Rice parameter m value, according to a data characteristic, freely selecting a block size, and raising an updating speed of the Golomb-Rice parameter m value; carrying out adaptive m value correction on the characteristic data to realize lossless audio frequency coding and decoding. The coding and decoding method of the present invention has the characteristics of low operation complexity and small storage space occupation, compression of an integer is realized, and especially a high compression ratio and a high coding and decoding speed of data with a geometric distribution characteristic are realized. The method is easily applied to the digital audio/video field and is suitable for realization with a modern processor, and processor cost of a user is substantially reduced.

Description

The back is to block adaptive Golomb-Rice decoding method and device

Technical field

The present invention relates to the DAB information source encoding and decoding technique of digital signal processing, is specifically related to the method and apparatus of back to block adaptive Golomb-Rice encoding and decoding.

Background technology

Follow the development of computer science and technology, digital storage technique is constantly development also, and the density of storage medium is more and more come high; Memory capacity is increasing, and the size of multi-medium data also constantly (for example: high-definition digital video, 3D video is increasing; High-quality digital audio; High-resolution digital images etc.), so compressing data is still essential, and requires compression and decompression to have lower algorithm complex.How to utilize effective method that raw data is expressed as bit still less, it is one of focus of present field of data compression research that the process of compression and decompression simultaneously keeps lower algorithm complex.

In order to realize the compression to data-signal, main method is the characteristic according to data-signal, removes space, time and the statistic correlation of data.Present various compression and decompression algorithm utilization prediction; What converter technique was removed is data space and time correlativity; Utilize entropy coding to remove the statistic correlation of data, main entropy coding has arithmetic coding, Huffman encoding; Run-length encoding, Columbus's coding (Golomb-Coding) etc.

Columbus's coding is a kind of important entropy coding of nineteen sixty by Solomon W.Golomb invention, for the alphabet of following geometric distributions, is a kind of optimum entropy coding method.The Golomb-Rice coding is the sub-set that Columbus encodes, and on Columbus's basis of coding, invents generation by Robert F.Rice, although compression effectiveness can be slightly poorer than Columbus's coding; Be more suitable for computer implemented advantage but have; Therefore application is very extensive, has simply characteristics of high efficiency.

The Golomb-Rice coding is applicable to the data source that meets geometric distributions is compressed; Do not consider the probability model of actual information source; Its compression performance generally speaking can be than arithmetic, and the Huffman entropy coding is a bit weaker, but because its computational complexity is low; It is applied to various compression field, as having adopted the Golomb-rice entropy encoding/decoding among the JPEG-LS; In lossless audio coding by extensively, the lossless audio compression standard ALS of MPEG for example, what all adopt among famous lossless audio technology FLAC and Monkey ' the s Audio is the Golomb-rice entropy encoding/decoding of Golomb-rice and mutation.

The Golomb-Rice entropy coding is to encode to signless round values n, reduces the number of bits that signless round values n uses, and the Golomb-Rice parameter m value according to selecting is divided into two parts with signless round values n: prefix

With remainder r=n& (2 ^m-1), prefix q is adopted monobasic sign indicating number coding (unary coding), remainder adopts m binary digit to encode.The Golomb-Rice entropy coding is the coding to single unsigned number, and representes the unsigned number that is encoded with q+1+r binary digit, and the coding of single unsigned number only possibly be an integer binary digit, and the above process of iteration is to complete the source symbol sequence of encoding.

The m value of storage perhaps was about to the m (adaptive approach) of the source symbol value of decoding when Golomb-Rice entropy decode procedure read coding from code stream according to decoded source symbol value dynamic calculation; Monobasic decodes prefix q; Read m binary digit and obtain remainder r, finally decode source symbol value n=q*2 ^m| r, the above process of iteration is to decoding whole source symbol sequence.

From the current application of using Golomb-Rice to carry out data compression, can find; Be delivered to the input value of Golomb-Rice coding, majority is through pretreated data, and main preprocess method respectively has difference according to using; In the DAB lossless compress; Many employing linear predictions remove the correlativity of source data in time domain, and the residual error data distribution majority after the processing meets the geometric distributions characteristic, and the residual error data after therefore handling is fit to utilize the Golomb-Rice coding to compress.When concrete use Golomb-Rice coding compressed, a key issue was that the Golomb-Rice parameter m is worth choosing, and how to choose the quality that optimum Golomb-Rice parameter m value will influence compression effectiveness.At Shorten, in the FLAC lossless audio compression, through being that the piece that input vector is divided into predetermined length is solved.For each piece, twice reading of data of scrambler.Read for the first time, calculate the mean value of input value, and confirm the m value of this piece according to mean value, the m value grammer field of this piece of bit output, and with encode each data of this piece of m value is exported the code word after each digital coding.This method is used in a lot of lossless compress systems, is called as " pressing block adaptive " perhaps " forward direction self-adaptation " model [2].The deficiency of this model is that at first cataloged procedure needs twice through data to be encoded, influences coding rate, secondly is a difficult point when selecting block size.

A kind of solution is to use the back to adaptive technique.Encoder is reached an agreement to the original state of each piece.Scrambler uses the rule of acquiescence to produce code element like this, and the rule of demoder utilization acquiescence oppositely recovers coded data.Utilize predetermined adaptation rule to upgrade codec parameters m value.Can effectively reduce the transmission expense of parameter like this.Although the back can solve some problems in the forward direction adaptive model to adaptive approach; But also there are some distinctive deficiencies in the back to self-adaptation: first problem: the back is preparatory statistics expense and the parameter transmission expense that reduces coding side through the computational complexity that increases decoding end to adaptive approach; When decoding end is upgraded the GolomB-Rice parameter through predetermined adaptation rule; Increase the certain calculation complexity, reduced decoding speed; Second problem: in the compressed source data,, have the lower problem of self-adaptation accuracy like this, can influence compressibility if exist special data can cause the fluctuation of partial data bigger.

Therefore, a kind of above argumentation forward direction and back of can combining has very strong practical value to the harmless Golomb-Rice decoding method of model advantage with device.

Summary of the invention

The objective of the invention is to adopt simple back to the Golomb-Rice of block adaptive strategy decoding method and device; In conjunction with forward direction self-adaptation and back to adaptive advantage; Utilize simple Forecasting Methodology to confirm initial m value of Golomb-Rice and auto-adaptive parameter; Adopt moving window that auto-adaptive parameter is upgraded, adopt block division method to upgrade the m value according to auto-adaptive parameter.

This method confirms that through the simple forecast method initial m value and auto-adaptive parameter are to improve accuracy and the adaptive speed of back to self-adaptive initial m value in encoder-side; And pass through the frequency that piecemeal upgrades the method reduction codec renewal m value of m value, thereby improve the speed of encoding and decoding.

The object of the invention is realized through following scheme:

The present invention proposes a kind of back to block adaptive Golomb-Rice decoding method, and this method is encoded to self-adaptation Golomb-Rice after utilizing the adaptive updates rule to realize;

Utilize the moving window method to upgrade auto-adaptive parameter;

Adopt piecemeal to upgrade to Golomb-Rice parameter m value,, freely select the branch block size, improve Golomb-Rice parameter m value renewal speed according to data characteristic;

Characteristic is carried out the correction of self-adaptation m value.

Said decoding method comprises concrete steps:

At coding side:

1) utilizes the method for forward prediction to calculate initial Golomb-rice parameter m value, calculate initial sum value Sum according to initial Golomb-rice parameter m value;

2) specify the branch block size according to data characteristic, utilize the input data of Golomb-rice parameter coding current block;

3) upgrade and value Sum according to the no sign map value of current block input data, utilize new and value Sum and 2 ^mThe magnitude relationship of * RICE_NUM_MUL is upgraded Golomb-rice parameter m value, and wherein, RICE_NUM_MUL is the length of moving window;

4) repeatable block Golomb-rice coding and piece Golomb-rice parameter update are to complete the input data of encoding;

In decoding end:

5) decode by the adaptation rule of agreement, at first the bit by the scrambler regulation reads out initial Golomb-rice parameter m value from bit stream, utilizes the initial sum value sum of Golomb-rice parameter m value initialization;

6) utilize the coded data of Golomb-rice parameter m value decoding current block;

7) after current block has been decoded, upgrade and value Sum according to the no sign map value of decoded data, and utilize new and value Sum and 2 ^mThe magnitude relationship of * RICE_NUM_MUL is upgraded the Golomb-rice parameter;

8) the Golomb-rice parameter update of repeatable block Golomb-rice decoding and piece is to complete the bit stream of decoding.

In step 1, the calculating of said initial Golomb-Rice parameter m value and initial sum value sum is following:

1. at coding side, at first adopt the relevant information of linear prediction method acquisition input data, and preceding L the data stronger according to correlativity, the mean value of the relevant progression length of calculating, its computing formula is following:

$\mathrm{mean}=\frac{\underset{i=0}{\overset{L}{\mathrm{\Σ}}}\left|\mathrm{data}\right[i\left]\right|}{L}---\left(1\right)$

2. calculate initial Golomb-Rice parameter m value, its computing formula is following:

$m\_\mathrm{init}=\left\{\begin{array}{cc}0& \mathrm{mean}<=1\\ \left(\mathrm{int}\right)\mathrm{floor}({\log }_2\left(\mathrm{mean}\right)+0.5)& \mathrm{mean}>1\end{array}\right.---\left(2\right)$

Export initial Golomb-Rice parameter m value with the long M of initial fixation bit;

3. utilize initial m value further to obtain initial sum value sum

The initial sum value is calculated as follows:

sum＝2 ^m×RICE_NUM_MUL (3)

Wherein, RICE_NUM_MUL is a moving window length, sets according to data characteristic.

In step 2, said Golomb-rice coding is divided into prefix one primitive encoding, interval bit and three parts of mantissa's coding, its process is following:

1. import the no symbolism of data, the input data x of input current block relevant position, its formula is following:

$\mathrm{ux}=\left\{\begin{array}{cc}2*x& x>=0\\ -2*x-1& x<0\end{array}\right.---\left(4\right)$

Wherein ux is the value of corresponding input data x after no sign map;

2. according to Golomb-rice parameter m value, calculate prefix, mantissa value, its computing formula is following:

r＝ux-q*2 ^m

Utilize monobasic sign indicating number coding prefix q value,, repeat 1 with m bits of encoded mantissa value), 2) to having encoded current block.

In step 3, the renewal of said Golomb-rice parameter m value comprises:

When 2 ^m* RICE_NUM_MUL≤sum≤2 ^M+1During * RICE_NUM_MUL, the m value remains unchanged;

When sum＞2 ^M+1During * RICE_NUM_MUL, m increases, and the scope of increment is

When sum＜2 ^mDuring * RICE_NUM_MUL, the scope of decrement is

Wherein, the m value is that unit upgrades with the piece, and the size of piece is greater than 1 arbitrary integer less than RICE_NUM_MUL, and sum is RICE_NUM_MUL import a data and value, the more new formula of sum value:

$\mathrm{sum}=\mathrm{sum}+\underset{i=\mathrm{start}}{\overset{\mathrm{end}}{\mathrm{\&Sigma;}}}\left\{\begin{array}{cc}\left(\mathrm{data}\right[i]-\frac{\mathrm{sum}}{\mathrm{RICE}\_\mathrm{NUM}\_\mathrm{MUL}})& 0<=i<L\\ \left(\mathrm{data}\right[i]\&times;2-\frac{\mathrm{sum}}{\mathrm{RICE}\_\mathrm{NUM}\_\mathrm{MUL}})& L<=<\mathrm{length}\left(\mathrm{data}\right),\mathrm{residual}\left[i\right]>=0\\ (-\left(\mathrm{data}\right[i]\&times;2)-1)-1-\frac{\mathrm{sum}}{\mathrm{RICE}\_\mathrm{NUM}\_\mathrm{MUL}})& L<=i<\mathrm{length}\left(\mathrm{data}\right),\mathrm{residual}\left[i\right]<0\end{array}\right.---\left(7\right)$

The length of length (data) information source of indicating to encode wherein; Start is the initial index of piece of just encode, and end is the end index, start then, and the relation of end is:

blksize＝end-start+1 (8)

Wherein, blksize is the size of piece.

In step 5, the extraction of the initial Golomb-rice parameter m value of said Golomb-rice decoding processing and the calculating of initial sum value sum are following:

1. the extraction of Golomb-rice parameter m value is from bit stream, to read the m value by the M bit that scrambler is stipulated, its expression formula is following:

m＝getbits(M) (9)

Wherein the function of getbits is from bit stream, to read the M bit;

2. the calculating of initial sum value sum is following:

sum＝2 ^m×RICE_NUM_MUL (10)

Set RICE_NUM_MUL and be the fixed value of 2 whole power.

Also comprise correlativity of removing between the sound channel and the step of removing correlation of data in the sound channel in step 1,

Correlativity property between the said removal sound channel is made up of the decision-making of sound channel correlativity and sound channel decorrelation two parts:

The related coefficient that original channel is calculated in said sound channel decorrelation decision-making respectively with the related coefficient of difference coding sound channel, select whether to carry out sound channel decorrelation decision according to facies relationship numerical value, and whether adaptive adjustment sound channel does the threshold values of decorrelation;

Said sound channel decorrelation is that the result who handles according to sound channel decorrelation decision-making does the decorrelation between the channel data;

Correlation of data is that general forecast method is removed correlation of data in the sound channel in the said removal sound channel, employing be the disposal route of LPC linear prediction, it is made up of with LPC linearity decorrelation two parts compute optimal LPC exponent number and coefficient.

Said compute optimal LPC exponent number and coefficient adopt the Levinson-Durbin iterative algorithm;

The linear decorrelation of said LPC is the LPC exponent number that utilizes optimum, and coefficient and input data are removed the correlativity between the data in the sound channel, obtain the residual values corresponding to the input data, are the list entries x of N for length, and the formula of LPC generation residual sequence d is following:

Wherein x is the data after the decorrelation of falling tone road, and lpc_order is the optimum exponent number of LPC, and what c [] [] preserved is optimum LPC coefficient;

In step 8, comprise and recover correlation of data and the step of rebuilding harmless original channel information in the sound channel:

Correlation of data is according to the data of back after the residual values behind the Golomb-rice of the block adaptive coding and LPC exponent number recover pre-processing module with coefficient in the said recovery sound channel, recovers the interior correlation of data of sound channel.Formula by residual sequence d reconstruct LPC list entries x is:

Wherein lpc_order is the optimum exponent number of LPC, and what c [] [] preserved is optimum LPC coefficient;

According to output sequence x and have or not decorrelation sign between sound channel to recover the correlativity between the sound channel:

The harmless original channel information of said reconstruction is the round-off error that produces when utilizing comprise in the difference sound channel strange/calculate when even information is recovered to encode and sound channel, through following method:

L＝Mid+(Side+Side&1)/2；

R＝Mid-(Side-Side&1)/2；

Other sound channel to method for reconstructing with L and R, the final harmless original audio input signal that recovers.

The concrete process of said sound channel decorrelation is following:

To monophony Mono input, decorrelation between saying nothing

Stereo sound Stereo input, the decorrelation method is following between sound channel:

With sound channel Mid=(L+R)/2

Difference sound channel Side=(L-R)

Promptly convert the left and right acoustic channels input Mid and Side passage into and be sent to the next stage coding unit, then Mid and Side passage are carried out absolute coding;

To multichannel Multi channel input, the decorrelation method is following between sound channel:

If 1. L and R sound channel exist simultaneously, then L and R sound channel convert Mid and Side passage into;

If 2. LS and RS exist simultaneously, then LS and RS sound channel convert SMid and SSide passage into;

Wherein: SMid=(LS+RS)/2, SSide=LS-RS

If 3. LB and RB exist simultaneously, then LB and RB sound channel convert BMid and BSide passage into;

Wherein: BMid=(LB+RB)/2, BSide=LB-RB

If 4. the LFE sound channel exists, the LFE sound channel is left intact;

If 5. the C sound channel exists, the C sound channel is left intact.

The present invention also proposes a kind of back to block adaptive Golomb-Rice coding and decoding device, comprises encoder, and said scrambler comprises:

Pre-processing module; Constitute by the decision-making of sound channel correlativity and sound channel decorrelation two parts; Correlativity to sound channel centering is judged; The related coefficient of calculating original channel respectively with the related coefficient of difference coding sound channel, select whether to carry out sound channel decorrelation decision according to facies relationship numerical value, and whether adaptive adjustment sound channel does the threshold values of decorrelation;

The prediction processing module; Constitute by compute optimal LPC exponent number and coefficient and the linear decorrelation two parts of LPC; Adopt Levinson-Durbin iterative algorithm compute optimal LPC exponent number and coefficient, utilize optimum LPC exponent number, coefficient and input data; Remove the correlativity between the data in the sound channel, obtain residual values corresponding to the input data;

The back utilizes the further packed data of distribution character of data to block adaptive Golomb-rice coding module, improves the compression performance of scrambler;

The bitstream format module is pressed frame head, frame data, and the sound channel head, the hierarchical structure of channel data is organized code stream;

Said demoder comprises:

Bit diffluence formatting module decodes the characteristic information of data by the rule of said scrambler defined, and these characteristic informations will instruct the voice data behind demoder correct decoding compressed.

The back is to piece Self-adaptationThe Golomb-rice decoder module, carry out the inverse process of back to the Golomb-rice of block adaptive coding, recover the residual error data after the prediction processing;

The backward-predicted processing module according to the data of back after the residual values behind the Golomb-rice of the block adaptive coding and LPC exponent number recover pre-processing module with coefficient, is recovered correlation of data in the sound channel, the bit stream of input is carried out lossless compress decode;

Reverse pre-processing module, the output sequence x after handling according to inverse prediction and have or not that the decorrelation sign recovers the correlativity between the sound channel between sound channel, the final harmless original audio input signal that recovers.

Effect of the present invention

1) the present invention is in the balance of considering between entropy coding complexity and the encoding-decoding efficiency; A kind of the back to block adaptive Golomb-Rice decoding method of digital data compression that be applicable to is provided; It is low to have computational complexity, and storage space takies little, compares with other entropy coding methods with higher computational complexity; Utilize this method can realize compression, especially the data with geometric distributions characteristic are realized high compression rate and high encoding and decoding speed integer.

2) the inventive method has simply, and the characteristics that are easy to realize are easy to be applied to the digital audio/video field, are used for the lossless audio encoding and decoding.

3) decoding method that provides of the present invention is applicable in modern times and realizes on the processor, on the processor with fixed-point processing ability, can well realize, greatly reduces user's processor cost.

4) as the improving a little of lossless audio technology, can obtain application future in the autonomous lossless audio standard of China.

Description of drawings

Fig. 1 is the coding module structured flowchart of back of the present invention to block adaptive Golomb-Rice decoding method;

Fig. 2 is the decoder module structured flowchart of back of the present invention to block adaptive Golomb-Rice decoding method;

Fig. 3 is the coding process flow diagram of back of the present invention to block adaptive Golomb-Rice;

Fig. 4 is the coding process flow diagram of back of the present invention to block adaptive Golomb-Rice;

Fig. 5 is auto-adaptive parameter and m value renewal figure in the block adaptive Golomb-Rice decoding method of the present invention;

Fig. 6 is an encoding code stream storage format exemplary plot of the present invention;

Fig. 7 is lossless audio coding device applying examples figure;

Fig. 8 is lossless audio decoding device applying examples figure;

Fig. 9 is a multichannel decorrelation synoptic diagram;

Figure 10 is 2.1 sound channels (L|R|LFE) decorrelation synoptic diagram;

Figure 11 is 3.1 sound channels (L|R|C|LFE) decorrelation synoptic diagram;

Figure 12 is 5.1 sound channels (L|R|LS|RS|C|LFE) decorrelation synoptic diagram;

Figure 13 is 7.1 sound channels (L|R|Ls|Rs|Lb|Rb|C|LFE) decorrelation synoptic diagram.

Embodiment

For making the object of the invention, technical scheme and advantage clearer, below in conjunction with specific embodiment, to further explain of the present invention.

1. gordian technique of the present invention comprises:

(1) utilize simple adaptive updates rule realization back to encode to self-adaptation Golomb-Rice;

(2) utilize simple moving window method to upgrade auto-adaptive parameter;

(3) adopt piecemeal to upgrade to Golomb-Rice parameter m value, can divide block size freely to select, improved Golomb-Rice parameter m value renewal speed according to data characteristic;

(4) characteristic is carried out the correction of self-adaptation m value;

1.1, the back to block adaptive Golomb-rice decoding method

Back theing contents are as follows to the self-adaptation Golomb-rice of piecemeal decoding method:

At coding side, like Fig. 3, utilize the method for forward prediction to calculate initial Golomb-rice parameter m value, calculate initial sum value Sum according to initial Golomb-rice parameter m value; Specify to divide block size according to data characteristic afterwards, utilize the input data of Golomb-rice parameter coding current block, after current block encode, according to the no sign map value renewal of current block input data be worth Sum, utilize new and value Sum ' and 2 ^mThe magnitude relationship of * RICE_NUM_MUL is upgraded Golomb-rice parameter m value; Repeatable block Golomb-rice coding and piece Golomb-rice parameter update are to complete the input data of encoding afterwards.(RICE_NUM_MUL is the length (Fig. 5) of moving window mentioned above, also promptly predicts the number of the input data of having encoded of next piecemeal).

In decoding end,, decode by the adaptation rule of agreement like Fig. 4.At first the bit by the scrambler regulation reads out initial Golomb-rice parameter m value from bit stream; Utilize Golomb-rice parameter m value to come initialization initial sum value Sum; Utilize the coded data of Golomb-rice parameter m value decoding current block afterwards; After current block has been decoded, upgrade and value Sum according to the no sign map value of decoded data, and utilize new and value Sum and 2 ^mThe magnitude relationship of * RICE_NUM_MUL is upgraded the Golomb-rice parameter, and the Golomb-rice parameter update of repeatable block Golomb-rice decoding afterwards and piece is to complete the bit stream of decoding.

In the encoding and decoding end, initial Golomb-rice parameter m value, initial sum value Sum and required being consistent of piece Golomb-rice parameter value renewal process, decoding just can recover original input data like this.

The detailed description that encoding and decoding are handled is following:

1.1.1, encoding process

Encoding process mainly is divided into the calculating of initial Golomb-rice parameter m value and initial sum value sum, three parts of renewal of the Golomb-rice coding of piece and the Golomb-rice parameter m value of piece:

1. the calculating of initial Golomb-Rice parameter m value and initial sum value sum

1) at coding side, at first adopt the relevant information of Forecasting Methodology (for example linear prediction) acquisition input data, and according to

Preceding L the data that correlativity is stronger.Calculate the mean value of relevant progression length, its computing formula is following:

$\mathrm{mean}=\frac{\underset{i=0}{\overset{L}{\mathrm{\&Sigma;}}}\left|\mathrm{data}\right[i\left]\right|}{L}---\left(1\right)$

2) calculate initial Golomb-Rice parameter m value, its computing formula is following:

$m\_\mathrm{init}=\left\{\begin{array}{cc}0& \mathrm{mean}<=1\\ \left(\mathrm{int}\right)\mathrm{floor}({\log }_2\left(\mathrm{mean}\right)+0.5)& \mathrm{mean}>1\end{array}\right.---\left(2\right)$

Export initial Golomb-Rice parameter m value, wherein LN2=0.69314718055994529 with the long M of initial fixation bit

Utilize initial m value further to obtain the initial sum value

The initial sum value is calculated as follows:

sum＝2 ^m×RICE_NUM_MUL (3)

Wherein RICE_NUM_MUL is a moving window length, can set according to data characteristic, required length less than information source; In order to make decoding method simple, and avoid division arithmetic, can be set at the fixed value of 2 whole power, for example 32.

2. the Golomb-rice of piece encodes

The Golomb-rice coding is divided into prefix one primitive encoding, interval bit (1 bit) and three parts of mantissa's coding, and its process is following:

1) the no symbolism of input data is x (the input data of current block relevant position) like the input data, and its formula is following:

$\mathrm{ux}=\left\{\begin{array}{cc}2*x& x>=0\\ -2*x-1& x<0\end{array}\right.---\left(4\right)$

Wherein ux is the value of corresponding input data x after no sign map.

2) according to Golomb-rice parameter m value, calculate prefix, mantissa value, its computing formula is following:

r＝ux-q*2 ^m

Utilize monobasic sign indicating number coding prefix q value,, repeat 1 with m bits of encoded mantissa value), 2) to having encoded current block.

3. the renewal of the Golomb-rice parameter m value of piece

The m value is that unit upgrades with the piece, and the size of piece can be greater than 1 arbitrary integer less than RICE_NUM_MUL, for fear of division arithmetic, can select the number of 2 whole power.Update method be import data with RICE_NUM_MUL of moving window size with value and 2 ^mThe relation of * RICE_NUM_MUL value increases or reduces the m value.When 2 ^m* RICE_NUM_MUL≤sum≤2 ^M+1During * RICE_NUM_MUL, the m value remains unchanged; When sum＞2 ^M+1During * RICE_NUM_MUL, m increases, and the scope of increment is

When sum＜2 ^mDuring * RICE_NUM_MUL, the scope of decrement is

Increment or decrement were got 1 o'clock, and the adaptive speed of m value is slower, when getting maximal value, were actually with current and calculated the m value of sub-piece down with value.The example formula that following m value is upgraded:

The more new formula of sum value:

$\mathrm{sum}=\mathrm{sum}+\underset{i=\mathrm{start}}{\overset{\mathrm{end}}{\mathrm{\&Sigma;}}}\left\{\begin{array}{cc}\left(\mathrm{data}\right[i]-\frac{\mathrm{sum}}{\mathrm{RICE}\_\mathrm{NUM}\_\mathrm{MUL}})& 0<=i<L\\ \left(\mathrm{data}\right[i]\&times;2-\frac{\mathrm{sum}}{\mathrm{RICE}\_\mathrm{NUM}\_\mathrm{MUL}})& L<=<\mathrm{length}\left(\mathrm{data}\right),\mathrm{residual}\left[i\right]>=0\\ (-\left(\mathrm{data}\right[i]\&times;2)-1)-1-\frac{\mathrm{sum}}{\mathrm{RICE}\_\mathrm{NUM}\_\mathrm{MUL}})& L<=i<\mathrm{length}\left(\mathrm{data}\right),\mathrm{residual}\left[i\right]<0\end{array}\right.---\left(7\right)$

The length of length (data) information source of indicating to encode wherein; RICE_NUM_MUL is a moving window length, and start is the initial index of piece of just encode, and end is the end index, is blksize like the size of piece, start then, and the relation of end does

blksize＝end-start+1 (8)

1.1.2, the Golomb-rice decoding processing

1. extract initial m value

This module is divided into two parts: 1, the extraction of initial Golomb-rice parameter m value;

2, the calculating of initial sum value sum

The extraction of Golomb-rice parameter m value is from bit stream, to read the m value by the M bit that scrambler is stipulated, its expression formula is following:

m＝getbits(M) (9)

Wherein the function of getbits is from bit stream, to read the M bit.

The calculating of initial sum value sum is following:

sum＝2 ^m×RICE_NUM_MUL (10)

2. the Golomb-rice of piece decoding

According to Golomb-rice parameter m value, decode each data of current piecemeal, and calculate current piecemeal and value.

3. the renewal of the Golomb-rice parameter m value of piece

Utilize the moving window length data and value and current piecemeal and value, use the method identical with coding side to upgrade, Golomb-rice parameter m value repeated for 2～3 steps up to decoding the original source sequence.

In describing below of the present invention, accompanying drawing has been carried out reference, accompanying drawing constitutes a part of the present invention, but has wherein provided the particular example of embodiment of the present invention as explanation.

Explain implementation method of the present invention through a concrete example below.

Self-adaptation Golomb-Rice codec method disclosed herein can be used for multiple compression applications.The lossless audio residual error data that for example, can be used for the process prediction processing.

Fig. 1 and Fig. 2 are block diagrams, show back exemplary realization to block adaptive Golomb-Rice decoding method disclosed herein.Among Fig. 1, provide the block diagram of encoder section, among Fig. 2, provided the block diagram of demoder.But Fig. 1 and Fig. 2 just wherein can realize and use two kinds in the several different methods of the present invention.

With reference to figure 1, the back can co-exist in the enterprising enforcement usefulness of same computer equipment to block adaptive Golomb-Rice scrambler with reference to figure 2 backs to block adaptive Golomb-Rice demoder, also can independently exist in different computer equipments.The code stream that the scrambler that shows with reference to figure 1 produces can be decoded with reference to the demoder among the figure 2.

With reference to figure 7; Fig. 8. the present invention provides concrete back to block adaptive Golomb-Rice Code And Decode method use-case; Method among the present invention can be used in the lossless audio coding technology; The method of in the entropy coding module, using the present invention to provide can effectively improve compressibility, improves Code And Decode speed.

Introduce although this enforcement is example with the lossless audio, this application is that the present invention supports one of multiple application, and the method for introducing among the present invention is equally applicable to the compression of other integer datas.

[embodiment]

The realization of coding method and device

As shown in Figure 7, scrambler of the present invention is by pre-processing module, the prediction processing module, and the back is to block adaptive Golomb-rice coding module, and the bitstream format module constitutes, and is used for the sound signal of input is carried out lossless compression-encoding.Below through the detailed description of each module the principle of work of scrambler is described.

Pre-processing module:

The effect of pre-processing module is the correlativity of removing between the sound channel, reduces the redundant information between the sound channel, thereby improves the compression performance of scrambler.It is made up of the decision-making of sound channel correlativity and sound channel decorrelation two parts.

Sound channel decorrelation decision-making need be judged the correlativity of sound channel centering; The related coefficient of calculating original channel respectively with the related coefficient of difference coding sound channel, selects whether to carry out sound channel decorrelation decision according to facies relationship numerical value and enable, and whether adaptive adjustment sound channel does the threshold values of decorrelation; When correlativity is strong between sound channel; Reduce the threshold values of decorrelation, more import data and will do the decorrelation processing, when the correlativity between the sound channel is more weak; Improve the threshold values of decorrelation, more import data and will not do the decorrelation processing.

The result that the sound channel decorrelation is handled according to sound channel decorrelation decision-making does the decorrelation between the channel data, and its concrete process is following:

1) to monophony (Mono) input, decorrelation between saying nothing

2) stereo sound (Stereo) input, the decorrelation method is following between sound channel:

And sound channel (Mid)=(L+R)/2

Difference sound channel (Side)=(L-R)

Promptly convert the left and right acoustic channels input Mid and Side passage into and be sent to the next stage coding unit, then Mid and Side passage are carried out absolute coding.Fig. 9 is a synoptic diagram.

3) to multichannel (Multi channel) input, the decorrelation method is following between sound channel:

If 1. L and R sound channel exist simultaneously, then L and R sound channel convert Mid and Side passage into;

If 2. LS and RS exist simultaneously, then Ls and Rs sound channel convert SMid and SSide passage into;

Wherein:

SMid＝(LS+RS)/2

SSide＝LS-RS

If 3. LB and RB exist simultaneously, then LB and RB sound channel convert BMid and BSide passage into;

Wherein:

BMid＝(LB+RB)/2

BSide＝LB-RB

If 4. the LFE sound channel exists, the LFE sound channel is left intact;

If 5. the C sound channel exists, the C sound channel is left intact.

Figure 10, Figure 11, Figure 12, Figure 13 are decorrelation synoptic diagram between the sound channel of several kinds of typical sound channel modes.

Prediction module:

The effect of prediction module is that general forecast method is removed correlation of data in the sound channel, reduces the redundant information between the data in the sound channel, thereby improves the compression performance of scrambler.What adopt in the present embodiment is the disposal route of LPC linear prediction, and it is made up of compute optimal LPC exponent number and coefficient and the linear decorrelation two parts of LPC.

For compute optimal LPC exponent number and coefficient, the Levinson-Durbin iterative algorithm of employing.

The linear decorrelation of LPC is to utilize optimum LPC exponent number; Coefficient and input data; Remove the correlativity between the data in the sound channel, obtain the residual values corresponding to the input data, its amplitude is generally less than the input data before the LPC decorrelation; For length is the list entries x (x is the data after the decorrelation of falling tone road) of N, and the formula of LPC generation residual sequence d is following:

Wherein lpc_order is the optimum exponent number of LPC, and what c [] [] preserved is optimum LPC coefficient.

Back to the Golomb-rice of block adaptive coding module:

The back is to piece Self-adaptationThe effect of Golomb-rice coding module utilize the further packed data of distribution character of data, thereby improve the compression performance of scrambler.Process that it is concrete such as chapters and sections 1.1.1, encoding process are described, and the size of limit slippage window is 32, and the size of piecemeal is 2,4,8,16,32,

In order to handle the too big problem of number of the prefix zero that really causes by forecasting inaccuracy; The maximum number that in the Golomb-rice of piece coding, limits prefix zero is 63; When the number of prefix zero greater than 63 the time; Increase Golomb-rice coding parameter m up to the number of prefix zero smaller or equal to 63, and export the increment of m value behind the bit of interval.

Bitstream formatization

The effect of bitstream formatization is by frame head, frame data, and the sound channel head, the hierarchical structure of channel data is organized code stream, makes it be more suitable for the application in multimedia audio field.

The realization of coding/decoding method and device

As shown in Figure 8, demoder of the present invention is by bit diffluence format, and the back is to piece Self-adaptationGolomb-rice decoding, the backward-predicted processing module, reverse pre-processing module constitutes, and is used for the bit stream of input is carried out the lossless compress decoding.Below through the detailed description of each module the principle of work of demoder is described.

Bit diffluence format

The formative effect of bit diffluence decodes the characteristic information of data by the rule of scrambler defined, and these characteristic informations will instruct the voice data behind demoder correct decoding compressed.

The back is to the Golomb-rice of block adaptive decoding

The back is to piece Self-adaptationThe effect of Golomb-rice decoding be the residual error data that recovers after the prediction processing, it is the inverse process of back to the Golomb-rice of block adaptive coding, process that it is concrete such as chapters and sections 1.1.2, decoding processing are described; And the size of limit slippage window is 32; The size of piecemeal is 2,4,8; 16,32.

Wherein ricesum be preceding RiCE_NUM_MUL=32 coded data with, subblksum be the current sub-block coded data with, m is the used m value parameter of Golomb-rice coding current sub-block.

The backward-predicted processing module

The effect of backward-predicted processing module is according to the data after afterwards residual values after the Golomb-rice of block adaptive encodes and LPC exponent number and coefficient recover pre-processing module, recovers correlation of data in the sound channel.Formula by residual sequence d reconstruct LPC list entries x is:

Wherein lpc_order is the optimum exponent number of LPC, and what c [] [] preserved is optimum LPC coefficient.

Reverse pre-processing module

Output sequence x after the effect of reverse pre-processing module is handled according to inverse prediction and have or not that the decorrelation sign recovers the correlativity between the sound channel between sound channel, the final harmless original audio input signal that recovers.Its concrete process is following:

The round-off error that produces when utilizing comprise in the difference sound channel strange/calculate when even information is recovered to encode and sound channel.Realize rebuilding harmless original channel information through following method.

L＝Mid+(Side+Side&1)/2；

R＝Mid-(Side-Side&1)/2；

Other sound channel to method for reconstructing with L and R.

Above-described specific embodiment; The object of the invention, technical scheme and beneficial effect have been carried out further detailed description, and institute it should be understood that the above is merely specific embodiment of the present invention; Be not limited to the present invention; All within spirit of the present invention and principle, any modification of being made, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (9)

1. After one kind to block adaptive Golomb-Rice decoding method, it is characterized in that,

   Encode to self-adaptation Golomb-Rice after utilizing the adaptive updates rule to realize;

   Utilize the moving window method to upgrade auto-adaptive parameter;

   Adopt piecemeal to upgrade to Golomb-Rice parameter m value,, freely select the branch block size, improve Golomb-Rice parameter m value renewal speed according to data characteristic;

   Characteristic is carried out the correction of self-adaptation m value.
2. 2. method according to claim 1 is characterized in that said decoding method comprises concrete steps:

   At coding side:

   1) utilizes the method for forward prediction to calculate initial Golomb-rice parameter m value, calculate initial sum value Sum according to initial Golomb-rice parameter m value;

   2) specify the branch block size according to data characteristic, utilize the input data of Golomb-rice parameter coding current block;

   3) upgrade and value Sum according to the no sign map value of current block input data, utilize new and value Sum and 2 ^mThe magnitude relationship of * RICE_NUM_MUL is upgraded Golomb-rice parameter m value, and wherein, RICE_NUM_MUL is the length of moving window;

   4) repeatable block Golomb-rice coding and piece Golomb-rice parameter update are to complete the input data of encoding;

   In decoding end:

   5) decode by the adaptation rule of agreement, at first the bit by the scrambler regulation reads out initial Golomb-rice parameter m value from bit stream, utilizes the initial sum value sum of Golomb-rice parameter m value initialization;

   6) utilize the coded data of Golomb-rice parameter m value decoding current block;

   7) after current block has been decoded, upgrade and value Sum according to the no sign map value of decoded data, and utilize new and value Sum and 2 ^mThe magnitude relationship of * RICE_NUM_MUL is upgraded the Golomb-rice parameter;

   8) the Golomb-rice parameter update of repeatable block Golomb-rice decoding and piece is to complete the bit stream of decoding.
3. 3. method according to claim 2 is characterized in that, in step 1, the calculating of said initial Golomb-Rice parameter m value and initial sum value sum is following:

   1. at coding side, at first adopt the relevant information of linear prediction method acquisition input data, and preceding L the data stronger according to correlativity, the mean value of the relevant progression length of calculating, its computing formula is following:

   $\mathrm{mean}=\frac{\underset{i=0}{\overset{L}{\mathrm{\&Sigma;}}}\left|\mathrm{data}\right[i\left]\right|}{L}---\left(1\right)$

   2. calculate initial Golomb-Rice parameter m value, its computing formula is following:

   $m\_\mathrm{init}=\left\{\begin{array}{cc}0& \mathrm{mean}<=1\\ \left(\mathrm{int}\right)\mathrm{floor}({\log }_2\left(\mathrm{mean}\right)+0.5)& \mathrm{mean}>1\end{array}\right.---\left(2\right)$

   Export initial Golomb-Rice parameter m value with the long M of initial fixation bit.

   3. utilize initial m value further to obtain initial sum value sum

   The initial sum value is calculated as follows:

   sum＝2 ^m×RICE_NUM_MUL (3)

   Wherein, RICE_NUM_MUL is a moving window length, sets according to data characteristic.
4. 4. method according to claim 3 is characterized in that, in step 2, said Golomb-rice coding is divided into prefix one primitive encoding, interval bit and three parts of mantissa's coding, and its process is following:

   1. import the no symbolism of data, the input data x of input current block relevant position, its formula is following:

   $\mathrm{ux}=\left\{\begin{array}{cc}2*x& x>=0\\ -2*x-1& x<0\end{array}\right.---\left(4\right)$

   Wherein ux is the value of corresponding input data x after no sign map;

   2. according to Golomb-rice parameter m value, calculate prefix, mantissa value, its computing formula is following:

   

   r＝ux-q*2 ^m

   Utilize monobasic sign indicating number coding prefix q value,, repeat 1 with m bits of encoded mantissa value), 2) to having encoded current block.
5. 5. method according to claim 4 is characterized in that, in step 3, the renewal of said Golomb-rice parameter m value comprises:

   When 2 ^m* RICE_NUM_MUL≤sum≤2 ^M+1During * RICE_NUM_MUL, the m value remains unchanged;

   When sum＞2 ^M+1During * RICE_NUM_MUL, m increases, and the scope of increment is

   

   When sum＜2 ^mDuring * RICE_NUM_MUL, the scope of decrement is

   Wherein, the m value is that unit upgrades with the piece, and the size of piece is greater than 1 arbitrary integer less than RICE_NUM_MUL, and sum is RICE_NUM_MUL import a data and value, the more new formula of sum value:

   $\mathrm{sum}=\mathrm{sum}+\underset{i=\mathrm{start}}{\overset{\mathrm{end}}{\mathrm{\&Sigma;}}}\left\{\begin{array}{cc}\left(\mathrm{data}\right[i]-\frac{\mathrm{sum}}{\mathrm{RICE}\_\mathrm{NUM}\_\mathrm{MUL}})& 0<=i<L\\ \left(\mathrm{data}\right[i]\&times;2-\frac{\mathrm{sum}}{\mathrm{RICE}\_\mathrm{NUM}\_\mathrm{MUL}})& L<=<\mathrm{length}\left(\mathrm{data}\right),\mathrm{residual}\left[i\right]>=0\\ (-\left(\mathrm{data}\right[i]\&times;2)-1)-1-\frac{\mathrm{sum}}{\mathrm{RICE}\_\mathrm{NUM}\_\mathrm{MUL}})& L<=i<\mathrm{length}\left(\mathrm{data}\right),\mathrm{residual}\left[i\right]<0\end{array}\right.---\left(7\right)$

   The length of length (data) information source of indicating to encode wherein; Start is the initial index of piece of just encode, and end is the end index, start then, and the relation of end is:

   blksize＝end-start+1 (8)

   Wherein, blksize is the size of piece.
6. 6. method according to claim 5 is characterized in that, in step 5, the extraction of the initial Golomb-rice parameter m value of said Golomb-rice decoding processing and the calculating of initial sum value sum are following:

   1. the extraction of Golomb-rice parameter m value is from bit stream, to read the m value by the M bit that scrambler is stipulated, its expression formula is following:

   m＝getbits(M) (9)

   Wherein the function of getbits is from bit stream, to read the M bit;

   2. the calculating of initial sum value sum is following:

   sum＝2 ^m×RICE_NUM_MUL (10)

   Set RICE_NUM_MUL and be the fixed value of 2 whole power.
7. 7. method according to claim 1 is characterized in that, also comprises correlativity of removing between the sound channel and the step of removing correlation of data in the sound channel in step 1,

   Correlativity property between the said removal sound channel is made up of the decision-making of sound channel correlativity and sound channel decorrelation two parts:

   The related coefficient that original channel is calculated in said sound channel decorrelation decision-making respectively with the related coefficient of difference coding sound channel, select whether to carry out sound channel decorrelation decision according to facies relationship numerical value, and whether adaptive adjustment sound channel does the threshold values of decorrelation;

   Said sound channel decorrelation is that the result who handles according to sound channel decorrelation decision-making does the decorrelation between the channel data;

   Correlation of data is that general forecast method is removed correlation of data in the sound channel in the said removal sound channel, employing be the disposal route of LPC linear prediction, it is made up of with LPC linearity decorrelation two parts compute optimal LPC exponent number and coefficient.

   Said compute optimal LPC exponent number and coefficient adopt the Levinson-Durbin iterative algorithm;

   The linear decorrelation of said LPC is the LPC exponent number that utilizes optimum, and coefficient and input data are removed the correlativity between the data in the sound channel, obtain the residual values corresponding to the input data, are the list entries x of N for length, and the formula of LPC generation residual sequence d is following:

   

   Wherein x is the data after the decorrelation of falling tone road, and lpc_order is the optimum exponent number of LPC, and what c [] [] preserved is optimum LPC coefficient;

   In step 8, comprise and recover correlation of data and the step of rebuilding harmless original channel information in the sound channel:

   Correlation of data is according to the data of back after the residual values behind the Golomb-rice of the block adaptive coding and LPC exponent number recover pre-processing module with coefficient in the said recovery sound channel, recovers the interior correlation of data of sound channel.Formula by residual sequence d reconstruct LPC list entries x is:

   

   Wherein lpc_order is the optimum exponent number of LPC, and what c [] [] preserved is optimum LPC coefficient;

   According to output sequence x and have or not decorrelation sign between sound channel to recover the correlativity between the sound channel:

   The harmless original channel information of said reconstruction is the round-off error that produces when utilizing comprise in the difference sound channel strange/calculate when even information is recovered to encode and sound channel, through following method:

   L＝Mid+(Side+Side&1)/2；

   R＝Mid-(Side-Side&1)/2；

   Other sound channel to method for reconstructing with L and R, the final harmless original audio input signal that recovers.
8. 8. method according to claim 7 is characterized in that, the concrete process of said sound channel decorrelation is following:

   To monophony Mono input, decorrelation between saying nothing

   Stereo sound Stereo input, the decorrelation method is following between sound channel:

   With sound channel Mid=(L+R)/2

   Difference sound channel Side=(L-R)

   Promptly convert the left and right acoustic channels input Mid and Side passage into and be sent to the next stage coding unit, then Mid and Side passage are carried out absolute coding;

   To multichannel Multi channel input, the decorrelation method is following between sound channel:

   If 1. L and R sound channel exist simultaneously, then L and R sound channel convert Mid and Side passage into;

   If 2. LS and RS exist simultaneously, then LS and RS sound channel convert SMid and SSide passage into;

   Wherein: SMid=(LS+RS)/2, SSide=LS-RS

   If 3. LB and RB exist simultaneously, then LB and RB sound channel convert BMid and BSide passage into;

   Wherein: BMid=(LB+RB)/2, BSide=LB-RB

   If 4. the LFE sound channel exists, the LFE sound channel is left intact;

   If 5. the C sound channel exists, the C sound channel is left intact.
9. After one kind to block adaptive Golomb-Rice coding and decoding device, comprise encoder, it is characterized in that said scrambler comprises:

   Pre-processing module; Constitute by the decision-making of sound channel correlativity and sound channel decorrelation two parts; Correlativity to sound channel centering is judged; The related coefficient of calculating original channel respectively with the related coefficient of difference coding sound channel, select whether to carry out sound channel decorrelation decision according to facies relationship numerical value, and whether adaptive adjustment sound channel does the threshold values of decorrelation;

   The prediction processing module; Constitute by compute optimal LPC exponent number and coefficient and the linear decorrelation two parts of LPC; Adopt Levinson-Durbin iterative algorithm compute optimal LPC exponent number and coefficient, utilize optimum LPC exponent number, coefficient and input data; Remove the correlativity between the data in the sound channel, obtain residual values corresponding to the input data;

   The back utilizes the further packed data of distribution character of data to block adaptive Golomb-rice coding module, improves the compression performance of scrambler;

   The bitstream format module is pressed frame head, frame data, and the sound channel head, the hierarchical structure of channel data is organized code stream;

   Said demoder comprises:

   Bit diffluence formatting module decodes the characteristic information of data by the rule of said scrambler defined, and these characteristic informations will instruct the voice data behind demoder correct decoding compressed.

   The back is to piece Self-adaptationThe Golomb-rice decoder module, carry out the inverse process of back to the Golomb-rice of block adaptive coding, recover the residual error data after the prediction processing;

   The backward-predicted processing module according to the data of back after the residual values behind the Golomb-rice of the block adaptive coding and LPC exponent number recover pre-processing module with coefficient, is recovered correlation of data in the sound channel, the bit stream of input is carried out lossless compress decode;

   Reverse pre-processing module, the output sequence x after handling according to inverse prediction and have or not that the decorrelation sign recovers the correlativity between the sound channel between sound channel, the final harmless original audio input signal that recovers.

Images