CN102256133B - Distributed video coding and decoding method based on side information refining - Google Patents

Abstract

The invention discloses a distributed video coding and decoding method based on side information refining. A coding method comprises the following steps: wavelet transforming, Gray code coding and LDPC (Low Density Parity Check) coding. The coding method is characterized in that a WZ (Wyner-Ziv) frame is subject to sorting coding before the wavelet transforming. A decoding method comprises the following steps: LDPC decoding, Gray code decoding, wavelet inverse transforming and reconfiguring. The decoding method is technically characterized by further comprising a step of side information refining processing after reconfiguring, wherein the side information refining processing comprises the following steps of: acquiring initial side information by using a motion compensation weighting interpolating method, refining and updating the side information by using motion compensation of partially decoded WZ frame, and sending the refined side information to a decoder for decoding to acquire a motion compensation refined reconfigured frame. The method provided by the invention is reasonable in design. The WZ frame is subject to the sorting coding at a coding end and the motion compensation weighting interpolating and partially decoded WZ frame are used for promoting the quality of the side information at a decoding end, thereby promoting the property of reconfigured frame and promoting the rate distortion property of the distributed video coding under the condition of keeping low coding complexity.

Description

A kind of distributed video coding and coding/decoding method of refining based on side information

Technical field

The invention belongs to wireless video sensor network field, especially a kind of distributed video coding and coding/decoding method of refining based on side information.

Background technology

Distributed video coding (Distributed Video Coding, DVC) be a kind of brand-new Video Encoding Mode, it is based on Slepian-Wolf and Wyner-Ziv multiple source information encoding theory, adopt intraframe coding to add the interframe decoding, computation complexity is transferred to decoding end from coding side, have the characteristics such as encoder complexity is low, coding side power consumption is low, zmodem.Make it be particularly suitable for all limited wireless video terminals (as wireless video monitoring system, video sensor network etc.) of some computing capabilitys, memory size, power consumption, and the growth of using along with these new videos is got up in fast development in recent years.

In recent years increasing scholar, organize and begin DVC is furtherd investigate, and set up corresponding DVC application system.Typical DVC model comprises: Stamford model and Berkeley model.Wherein video sequence is divided into key frame and WZ frame usually.on this basis, two kinds of methods of existing broad research are respectively based on the pixel domain coding with based on transform domain coding, its difference is in preliminary treatment and reprocessing to the WZ frame, based on transform domain coding with respect to carrying out discrete cosine transform (Discrete Cosine Transform to video image based on pixel domain coding, DCT) or wavelet transform (Discrete Wavelet Transform, DWT) and inverse transformation, correspondingly encoder can utilize the spatial statistics correlation of video image, therefore its distortion performance is better than encoding based on pixel domain, but encoder complexity increases to some extent.In addition, because DWT can also overcome the blocking effect that the conversion of DCT piece brings, obtain simultaneously better Image Coding performance.

In the DVC system, the code efficiency of encoder largely depends on the reconstruction quality of side information.The side information reconstruction quality is higher, and the needed Parity Check Bits of decoding end decoding success is just fewer, and the compression efficiency of encoder is also just higher.Usually continuous key frame is carried out the motion-compensated temporal interpolation and obtain side information, still, the motion vector in the motion-compensated temporal interpolation usually is not consistent with the motion of actual object, so the reconstruction quality of side information is just not ideal enough.For this problem, Klomp, S. and Vatis, Y. have proposed a kind of side information generation technique that uses the sub-pixel precision of motion compensation, and Liu Rongke etc. propose a kind of side information generating algorithm based on Hierarchical Motion Estimation.Yet, do not consider decoded WZ frame in these technology, in fact the WZ frame of decoding includes than the more additional information of reference frame.In sum, existing distributive vide frequency coding method exists the ropy problem of side information and the low problem of distortion performance.

Summary of the invention

The object of the invention is to overcome the deficiencies in the prior art, propose a kind of distributed video coding and coding/decoding method of refining based on side information that can improve side information quality and distortion performance.

The present invention solves its technical problem and takes following technical scheme to realize:

A kind of distributive vide frequency coding method of refining based on side information comprises wavelet transformation, gray encoding and LDPC coding step, also is included in the WZ frame to be carried out the sorting code number step before wavelet transformation, and its concrete processing method is:

(1) calculate the absolute error of current macro and sad value and absolute error and the SAD of piece around all _nValue;

(2) absolute error and the SAD of piece around the basis _nInformation source is classified: if SAD _nValue is less than the threshold value of setting, and thinks that the piece correlation of relevant position is strong in current block and side information, and defining this piece is the skip pattern, if SAD _nValue is greater than the threshold value of setting, and thinks that a little less than the piece correlation of relevant position in current block and side information, current block is defined as inter-frame mode.

And the absolute error of described current macro and sad value adopt following formula to calculate:

$\mathrm{SAD}=\underset{x=1}{\overset{M}{\mathrm{\Σ}}}\underset{y=1}{\overset{N}{\mathrm{\Σ}}}|B(x,y)-B_{\mathrm{re}}(x,y)|$

In formula, B represents current block, B _reThe piece identical with the current block position in the expression reference frame, M, N represent block size;

Described absolute error and the SAD of piece around all _nValue adopts following formula to calculate:

${\mathrm{SAD}}_n=\frac{({\mathrm{<figure-callout\; id="0"\; label="SAD"\; filenames="HDA0000086165680000022.png"\; state="\{\{state\}\}">SAD</figure-callout>}}_0+\mathrm{\&lambda;}\&CenterDot;{\mathrm{SAD}}_1)}{1+\mathrm{\&lambda;}}$

In formula, λ is weighted value, SAD ₀The SAD of expression current block, and SAD ₁The average sad value of piece around expression.

And described inter mode block is further classified, if B and B _reBetween the SAD of high frequency coefficient _nLower than the threshold value of setting, think that the estimated value of decoding end is enough accurately given up high frequency coefficient, otherwise will deliver to decoding end after their codings.

A kind of distributed video decoding of refining based on side information comprises LDPC decoding, Gray code decoding, inverse wavelet transform and reconstruction step, also comprises the side information treatment step of refining after reconstruct, and its concrete processing method is:

(1) adopt motion compensation Weighted Interpolation method to obtain initial side information;

(2) use the motion compensation of the partial decoding of h WZ frame renewal side information of refining;

(3) side information after refining is sent into decoder and is decoded and obtain the reconstruction frames that motion compensation is refined.

And described motion compensation Weighted Interpolation method comprises the steps:

(1) at key frame X ' _2i+2Middle search and former frame X ' _2iIn macro block S _2iThe macro block S that is complementary _2i+2, obtain corresponding motion vector MV _f

(2) with X ' _2iIn macro block S _2iBe reference, use MV _f/ 2 obtain interior inserted block B ₁₁, with X ' _2i+2In macro block S _2i+2Be reference, use-MV _f/ 2 obtain interior inserted block B ₁₂

(3) at key frame X ' _2iMiddle search and a rear frame X ' _2i+2In macro block S _2i+2The macro block S that is complementary _2i, obtain corresponding motion vector MV _bWith X ' _2i+2In macro block S _2i+2Be reference, use MV _b/ 2 obtain interior inserted block B ₂₁, with X ' _2iIn macro block S _2iBe reference, use-MV _b/ 2 obtain interior inserted block B ₂₂Thereby, obtain the initial side information after motion compensation:

${\mathrm{<figure-callout\; id="2i"\; label="y"\; filenames="HDA0000086165680000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_{2i+1}=\frac{(b_{11}+b_{12}+b_{21}+b_{22})}{4}$

In formula, y _2i+1, b ₁₁, b ₁₂, b ₂₁, b ₂₂Be respectively Y _2i+1, B ₁₁, B ₁₂, B ₂₁, B ₂₂Pixel value.

And the motion compensation of described use partial decoding of h WZ frame is refined and is upgraded the side information method and comprise the steps:

(1) at former frame X ' _2iThe WZ frame WZ of middle searching partial decoding of h _dMatch block, draw forward motion compensation frame Z _F, at a rear frame X ' _2i+2Middle searching WZ _dMatch block, draw reverse compensated frame Z _B

(2) calculate according to the following formula WZ _dWith Z _F, Z _BResidual error:

R _F(x，y)＝Z _F(x，y)-WZ _d(x，y)

R _B(x，y)＝Z _B(x，y)-WZ _d(x，y)

(3) according to the residual error size, calculate according to the following formula weight coefficient p:

$U=e^{-R_F{(x,y)}^2}$

$V=e^{-R_B{(x,y)}^2}$

$p=\frac{U}{U+V}$

(4) calculate according to the following formula side information after motion compensation is refined:

Y＝pZ _F+(1-p)Z _B。

Advantage of the present invention and good effect are:

The present invention is reasonable in design, at coding side, the WZ frame is carried out sorting code number, adopt the WZ frame of motion compensation Weighted Interpolation and partial decoding of h to improve the quality of side information in decoding end, and then the performance of raising reconstruction frames, the method can in the situation that encoder complexity is hanged down in maintenance, more effectively improve the distortion performance of distributed video coding.

Description of drawings

Fig. 1 is distributed video coding proposed by the invention and the processing procedure schematic diagram of decoding;

Fig. 2 is the generation schematic diagram of initial side information;

Fig. 3 is relatively schematic diagram of distortion performance;

Fig. 4 is that the subjective quality of the present embodiment Foreman sequence compares schematic diagram.

Embodiment

Below in conjunction with accompanying drawing, the embodiment of the present invention is further described:

A kind of distributed video coding and coding/decoding method of refining based on side information, as shown in Figure 1, increased WZ frame classification coding at coding side and processed the coding gain that is used for improving distributed video coding, refined at decoding end increase side information and process the quality that is used for improving side information.The below respectively to the distributive vide frequency coding method of coding side and and the distributed video coding/decoding method of decoding end describe.

A kind of distributive vide frequency coding method of refining based on side information comprises: the WZ frame is carried out sorting code number step, wavelet transformation step, gray encoding step and LDPC coding step, and wherein, it is innovation of the present invention that the WZ frame is carried out the sorting code number step.The core of WZ frame classification coding is to adopt new decision rule to carry out sorting code number to the WZ frame, and its concrete grammar is as follows:

For a given piece, adopt following formula to calculate absolute error and (SAD):

$\mathrm{SAD}=\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}|B(x,y)-B_{\mathrm{re}}(x,y)|---\left(1\right)$

Wherein, B represents current block, B _reThe piece identical with the current block position in the expression reference frame, M, N represents block size.

Consider time domain direction property and spatial coherence, adopt following formula to calculate the sad value of current macro and the SAD of piece around all _nValue, SAD _nDefinition is:

${\mathrm{SAD}}_n=\frac{({\mathrm{<figure-callout\; id="0"\; label="SAD"\; filenames="HDA0000086165680000022.png"\; state="\{\{state\}\}">SAD</figure-callout>}}_0+\mathrm{\&lambda;}\&CenterDot;{\mathrm{SAD}}_1)}{1+\mathrm{\&lambda;}}---\left(2\right)$

Wherein, λ is weighted value, SAD ₀The SAD of expression current block, and SAD ₁The average sad value of piece around expression.

According to SAD _{n value}Information source is classified, and concrete principle of classification is as follows:

If SAD _nValue is less than the threshold value of a certain setting, thinks that the piece correlation of relevant position is strong in current block and side information, and defining this piece is the skip pattern, directly replaces with the side information value of estimating in decoding end;

If SAD _nValue is greater than the threshold value of a certain setting, thinks that a little less than the piece correlation of relevant position in current block and side information, current block is defined as inter-frame mode, then carries out wavelet transformation.

In order to reduce the expression cost of relevant information, inter mode block can further be classified.If B and B _reBetween the SAD of high frequency coefficient _nLower than the threshold value of setting, enough accurately give up high frequency coefficient with regard to the estimated value of supposition decoding end, otherwise will deliver to decoding end after their codings.

Above-mentioned the WZ frame carried out the sorting code number method, block mode has reflected movable information, thereby it helps estimation and high frequency coefficient to recover.On the other hand, those skip mode blocks are thought constant and can be used as the constraints of LDPC coding in order to improve performance.It is better that the decoding end side information is estimated, skip mode block and the high frequency coefficient that can abandon are just more.Equally, block sort is more accurate, and the estimation of side information is also more accurate.Therefore, the classification of the information source of coding side can significantly improve coding gain.

A kind of distributed video coding/decoding method of refining based on side information comprises LDPC decoding step, Gray code decoding procedure, inverse wavelet transform step, reconstruction step and the side information treatment step of refining, and wherein, side information is refined and processed is innovation of the present invention.Directly affect the RD performance of whole encoder due to the reconstruction quality of side information in the distributed video coding device, the side information of decoding end is more accurate shows that to need the encoding error of error correction fewer, the check bit that need to send to decoding end is also fewer, when obtaining same decoding WZ frame quality, required bit rate has just reduced like this.Following opposite side Information refinement is processed and is elaborated:

(1) adopt motion compensation Weighted Interpolation method to obtain initial side information.

As shown in Figure 2, motion search is at key frame X ' _2iAnd X ' _2i+2Between carry out.The concrete steps of motion compensation Weighted Interpolation method are:

At first, at key frame X ' _2i+2Middle search and former frame X ' _2iIn macro block S _2iThe macro block S that is complementary _2i+2, obtain corresponding motion vector MV _fSecondly, with X ' _2iIn macro block S _2iBe reference, use MV _f/ 2 obtain interior inserted block B ₁₁, in like manner, with X ' _2i+2In macro block S _2i+2Be reference, use-MV _f/ 2 obtain interior inserted block B ₁₂At last, at key frame X ' _2iMiddle search and a rear frame X ' _2i+2In macro block S _2i+2The macro block S that is complementary _2i, obtain corresponding motion vector MV _bEqually, with X ' _2i+2In macro block S _2i+2Be reference, use MV _b/ 2 obtain interior inserted block B ₂₁, with X ' _2iIn macro block S _2iBe reference, use-MV _b/ 2 obtain interior inserted block B ₂₂Therefore, the initial value of side information is:

${\mathrm{<figure-callout\; id="2i"\; label="y"\; filenames="HDA0000086165680000021.png"\; state="\{\{state\}\}">y</figure-callout>}}_{2i+1}=\frac{(b_{11}+b_{12}+{\mathrm{<figure-callout\; id="21"\; label="b"\; filenames="HDA0000086165680000011.png,HDA0000086165680000021.png"\; state="\{\{state\}\}">b</figure-callout>}}_{21}+b_{22})}{4}---\left(3\right)$

Wherein, y _2i+1, b ₁₁, b ₁₂, b ₂₁, b ₂₂Be respectively Y _2i+1, B ₁₁, B ₁₂, B ₂₁, B ₂₂Pixel value.

(2) use the motion compensation of the partial decoding of h WZ frame renewal side information of refining

The purpose that motion compensation is refined is to obtain more accurate reconstruction frames from current decoding WZ frame.To produce the WZ frame of a partial decoding of h, i.e. WZ by the WZ decoder based on initial side information _d, and utilize WZ _dThe side information of refining.Its concrete steps are as described below:

At first, at former frame X ' _2iMiddle searching WZ _dMatch block, draw forward motion compensation frame Z _F, at a rear frame X ' _2i+2Middle searching WZ _dMatch block, draw reverse compensated frame Z _B

Secondly, calculate WZ _dWith Z _F, Z _BResidual error

R _F(x，y)＝Z _F(x，y)-WZ _d(x，y) (4)

R _B(x，y)＝Z _B(x，y)-WZ _d(x，y) (5)

Then, according to the residual error size, calculate weight coefficient

$U=e^{-R_F{(x,y)}^2}---\left(6\right)$

$V=e^{-R_B{(x,y)}^2}---\left(7\right)$

$p=\frac{U}{U+V}---\left(8\right)$

At last, calculate according to the following formula side information after motion compensation is refined:

Y＝pZ _F+(1-p)Z _B (9)

(3) side information after refining is sent into decoder and is decoded and obtain the reconstruction frames that motion compensation is refined, thereby has improved the distortion performance of whole system.

The below uses an example that the present invention is verified.The simulated environment that adopts in case verification is matlab.R2007b, emulation experiment respectively in selection standard video sequence storehouse the motion more rich Foreman sequence of more violent details and the lighter Hall sequence of motion severe degree test, these two video sequences all adopt QCIF (176 * 144) form, speed is 30fps, therefore the speed of odd-numbered frame is 15fps.Odd-numbered frame is encoded as the WZ frame and even frame is key frame, and each WZ frame is broken down into the macro block of 16x16, makes 3 grades of Daubechies ' 9-7 wavelet transformations.In decoding end, decoded even frame will be used for interpolation and generate the decoding that side information is used for odd-numbered frame, and decoding end error code thresholding is set as P _e=10 ^-3The present invention adopts the irregular LDPC codes (n=10000, k=5000) of 1/2 code check.The generator polynomial of LDPC code is:

λ(x)＝0.234029x+0.212425x ²+0.146898x ⁵+0.10284x ⁶+0.303808x ¹⁹ (10)

ρ(x)＝0.71875x ⁷+0.28125x ⁸ (11)

Whole rate distortion in experimentation (RD:Rate Distortion) curve is only considered average bit rate and average peak signal to noise ratio (PSNR) value of Wyner-Ziv frame luminance component, and for without loss of generality, test result is with H.264 intraframe coding and interframe encode contrast, wherein, H.264/JVTJM16 encoder encodes is adopted in the unification of Key frame.

Fig. 3 has shown the distortion performance of Foreman and Hall QCIF sequence.For the Hall sequence slowly of moving, adopting the side information refined method that systematic function is improved not is clearly; But for the more violent Foreman sequence of motion.The scheme that the present invention proposes is than traditional transform domain TDWZ encoding scheme, and population rate distortion performance maximum has improved the 2dB left and right.

Fig. 4 is the subjective quality comparative result of Foreman sequence.Wherein Fig. 4 (a) is original graph, and Fig. 4 (b) is motion compensated interpolation figure, and Fig. 4 (c) is the treatment effect figure after purifying through motion compensation, the reconstructed image better effects if after as can be seen from the figure motion compensation is refined.In Fig. 4 (b), there is obvious distortion in the part of nose and mouth.After refining by motion compensation, it is very clear that the moving region becomes.

It is emphasized that; embodiment of the present invention is illustrative; rather than determinate; therefore the present invention is not limited to the embodiment described in embodiment; every other execution modes that drawn by those skilled in the art's technical scheme according to the present invention belong to the scope of protection of the invention equally.

Claims (2)

1. a distributive vide frequency coding method of refining based on side information, comprise wavelet transformation, gray encoding and LDPC coding step, it is characterized in that: before also being included in wavelet transformation, the WZ frame is carried out the sorting code number step, its concrete processing method is:

⑴ calculate the absolute error of current macro and sad value and absolute error and the SAD of piece around all _nValue;

⑵ absolute error and the SAD of piece around basis _nInformation source is classified: if SAD _nValue is less than the threshold value of setting, and thinks that the piece correlation of relevant position is strong in current block and side information, and defining this piece is the skip pattern, if SAD _nValue is greater than the threshold value of setting, and thinks that a little less than the piece correlation of relevant position in current block and side information, current block is defined as inter-frame mode;

The absolute error of described current macro and sad value adopt following formula to calculate:

$\mathrm{SAD}=\underset{x=1}{\overset{M}{\mathrm{\&Sigma;}}}\underset{y=1}{\overset{N}{\mathrm{\&Sigma;}}}|B(x,y)-B_{\mathrm{re}}(x,y)|$

In formula, B represents current block, B _reThe piece identical with the current block position in the expression reference frame, M, N represent block size;

Described absolute error and the SAD of piece around all _nValue adopts following formula to calculate:

${\mathrm{SAD}}_n=\frac{({\mathrm{SAD}}_0+\mathrm{\&lambda;}\&CenterDot;{\mathrm{SAD}}_1)}{1+\mathrm{\&lambda;}}$

In formula, λ is weighted value, SAD ₀The SAD of expression current block, and SAD ₁The average sad value of piece around expression.

2. a kind of distributive vide frequency coding method of refining based on side information according to claim 1, it is characterized in that: described inter mode block is further classified, if B and B _reBetween the SAD of high frequency coefficient _nLower than the threshold value of setting, think that the estimated value of decoding end is enough accurately given up high frequency coefficient, otherwise will deliver to decoding end after their codings.

Images