CN102186077A - Wyner-Ziv-video-coding-based Wyner-Ziv frame code rate control system and method - Google Patents

Abstract

The invention discloses a Wyner-Ziv-video-coding-based Wyner-Ziv frame code rate control system and a Wyner-Ziv-video-coding-based Wyner-Ziv frame code rate control method, which belong to the technical field of distributed video coding and mainly aim to solve the problem that channel resources are not utilized rationally in the quantization matrix selection of a conventional Wyner-Ziv frame. The control system provided by the invention comprises a training module, a code rate estimation module and a selection module, wherein the training module is used for determining a band rate distortion model coefficient; the code rate estimation module calculates an estimated code rate by utilizing band rate-distortion of source-channel coding; and the selection module selects an optimal quantization matrix from eight candidate quantization matrixes. In the system and the method, the code rate of a coefficient band is estimated by establishing the band rate-distortion of source-channel coding, so the determined quantization matrix is more accurate, and the rate distortion performance of distributed video coding (DVC) is improved. A code rate allocation method provided by the invention can be used for feedback code rate control-free Wyner-Ziv video coding system.

Description

Wyner-Ziv frame Rate Control system and method based on the Wyner-Ziv video coding

Technical field

The invention belongs to technical field of video coding, relate to the distributed video coding system, specifically be a kind of Wyner-Ziv frame Rate Control system and method, can be used for actual Wyner-Ziv video coding system and the simple video communication system of terminal encoding device based on the Wyner-Ziv video coding.

Background technology

Along with the development of modern technologies, traditional video encoding standard H.264/AVC, MPEG-X can not satisfy some application scenario, such as wireless video sensor network, mobile video telephone and wireless video monitoring etc.Because using, these generally need the power consumption simple, that needs are little of encoding, so objectively need a kind of new coding/decoding system, this new coding/decoding system should have the formation of coding side and handle the characteristics of simplifying, the complexity of coding side is far smaller than the ratio decoder end, in other words the complexity of coding side is transferred to decoding end.

The distributed video coding technology (Distributed Video Coding, DVC) coding techniques with traditional is opposite, and it is the complicated a kind of new coding and decoding scheme of the simple decoding of coding.Diminish two theoretical foundations of source encoding based on Slepain-Wolf lossless coding theory and Wyner-Ziv, the distributed video coding technology reaches its maturity.Be used widely at wireless video sensor network, mobile video telephone and wireless video monitoring etc., therefore the Wyner-Ziv coded system also is a research focus of at present domestic and international distributed video coding, it is divided into key frame and Wyner-Ziv frame with video sequence, an image sets (Group of Pictures, GOP) comprise all Wyner-Ziv frames between a key frame and it and next key frame, it is convenient with GOP to be that unit encodes.

In distributed video coding, Rate Control is a very crucial technology, and it relates to the distortion performance of whole system.Wyner-Ziv frame Rate Control mainly comprises choosing and the bit-plane Rate Control of quantization matrix.By choosing mode with the quantization matrix of EI retrieval Wyner-Ziv frame, promptly how from alternative quantization matrix, to choose the optimum quantization matrix, find existing DVC system mostly according to determining quantization matrix with the empirical value of key frame quantization parameter coupling, the constraint with target bit rate does not take into account.The bit-plane Rate Control is divided into to be had the feedback Rate Control and not to have feedback Rate Control dual mode.Have feedback system to pass through feedback channel transfer check position repeatedly according to decoding end decoding situation, until decoding successfully, it is just can carry out under the online situation simultaneously that this mode needs encoding and decoding, and adjusting that can be unsuitable according to bandwidth requirement.Do not have the feedback code check and be controlled at coding side estimation code check to transmit suitable check digit, main correlation according to side information and source information.

In the superincumbent Data Rate Distribution scheme, when selecting quantization matrix, the simple principle suitable with the key frame quality according to the Wyner-Ziv frame, do not consider the characteristics of Wyner-Ziv frame combined channel source encoding, ignored of the influence of this coding characteristic to Wyner-Ziv frame code check, can not be according to the quantization matrix of choosing of Wyner-Ziv frame target bit rate change dynamics, finally can not the reasonable use channel resource, influence the decoding quality of video sequence.

Summary of the invention

The objective of the invention is to overcome the shortcoming of above-mentioned prior art, a kind of Wyner-Ziv frame Rate Control system and method based on the Wyner-Ziv video coding is proposed, by setting up coefficient tape rate-distortion model (Band Rate-Distortion of Source-Channel Coding in conjunction with signal source and channel, BRD-SCC), contrast target bit rate and eight estimation code checks that alternative quantization matrix is corresponding, selected code check near the time corresponding quantitative matrix quantization Wyner-Ziv frame, carry out the bit-plane Data Rate Distribution, improved the distortion performance of DVC system.

The present invention is a kind of Wyner-Ziv frame Rate Control system based on the Wyner-Ziv video coding, its be implemented in: this system includes: training module, the rate estimation module, choose module, training module receives first Wyner-Ziv frame of sequence to be encoded, utilize two typical quantization matrixes to train, obtain coefficient tape rate-distortion model coefficient in conjunction with signal source and channel, in the rate estimation module, utilize estimation code check in conjunction with eight quantization matrix correspondences of coefficient tape rate distortion BRD-SCC Model Calculation of signal source and channel, module is chosen in access, in choosing module, eight estimation code checks are compared with target bit rate, select to satisfy the quantization matrix of target bit rate.

At present, in distributed video coding system and the simple video communication system of terminal encoding device the Rate Control of Wyner-Ziv frame adopted and make the matching method that key frame is suitable with Wyner-Ziv frame quality, do not consider the influence of the bit-plane code check of Wyner-Ziv frame, finally can not the reasonable use channel resource.The present invention has overcome above-mentioned shortcoming, has proposed a kind of Wyner-Ziv frame Rate Control system of dynamic adjustment.

In whole Wyner-Ziv video coding system, after coding side carries out the distribution of frame layer bit rate, with the Wyner-Ziv frame target bit rate that obtains as input, access the present invention is based on the Wyner-Ziv frame Rate Control system of Wyner-Ziv video coding, reasonable use channel resource of the present invention, realize dynamic Wyner-Ziv frame Rate Control, output information of the present invention is as the input of uniform quantizer.

Realization of the present invention also is: two typical quantization matrixes that are used to train are

$Q1=\left[\begin{array}{cccc}7& 6& 5& 4\\ 6& 5& 4& 3\\ 5& 4& 3& 3\\ 4& 3& 3& 0\end{array}\right]$ $\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HSA00000486492200012.png"\; state="\{\{state\}\}">Q</figure-callout>}2=\left[\begin{array}{cccc}4& 3& 2& 2\\ 3& 2& 2& 2\\ 2& 2& 2& 2\\ 2& 2& 2& 0\end{array}\right]$

Element representation in the quantization matrix is used for the quantizing bit number of coefficient tape, numerical value is big more, quantize thin more, quantization step is more little, therefrom as can be seen: the quantification of Q1 is thinner, and the quantification of Q2 is thicker, and the rate distortion point that is obtained by these two matrixes differs far away, so the BRD-SCC model coefficient of determining is more accurate.

The present invention still is a kind of Wyner-Ziv frame bit rate control method based on the Wyner-Ziv video coding, and it is characterized in that: according to the selection quantization matrix of encoder bit rate change dynamics, the step of this control method includes:

Step 1, the Wyner-Ziv frame of choosing first GOP are carried out 4 * 4 discrete cosine transform DCT as training frames to it, and the extraction coefficient band;

Step 2, training utilizes two typical quantization matrix Q1 and Q2 to quantize the coefficient tape of training frames respectively, generates two rate distortion points, calculates BRD-SCC model coefficient a and b;

Step 3 reads current Wyner-Ziv frame, and it is carried out 4 * 4DCT, and the extraction coefficient band;

Step 4, select quantization matrix, BRD-SCC model coefficient a and b that usage factor band rate-distortion model and training obtain, calculate the estimation code check of eight quantization matrix correspondences, estimate that with eight code checks compare with the Wyner-Ziv frame target bit rate that the control of frame layer bit rate obtains, select and the pairing quantization matrix of the immediate estimation code check of target bit rate;

Step 5 is utilized the quantization matrix of selecting to obtain to quantize the Wyner-Ziv frame, and is extracted bit-plane;

Step 6 is carried out the bit-plane Data Rate Distribution according to the correlation of side information and source information;

Step 7, step 3 is to 6 being the Rate Control process of a Wyner-Ziv frame, repeating step 3 to 6 can be finished the Rate Control of all Wyner-Ziv frames of whole video sequence.

The present invention determines BRD-SCC model coefficient a and b by training process, this model is used to estimate each quantization matrix corresponding code rate, thereby the quantization matrix of selecting to satisfy target bit rate quantizes the Wyner-Ziv frame, because consider the constraint of target bit rate, realize the Wyner-Ziv frame Rate Control of video coding, so the present invention has more reasonably utilized channel resource.

Realization of the present invention also is: the BRD-SCC model to set up process as follows:

4.1, the pixel domain rate-distortion model of Wyner-Ziv frame:

$R=\left\{\begin{array}{cc}\frac{1}{2}{\log }_2\left(\frac{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="x"\; filenames="HSA00000486492200012.png"\; state="\{\{state\}\}">x</figure-callout>}}^2{\mathrm{\&sigma;}}_u^2}{({\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="x"\; filenames="HSA00000486492200012.png"\; state="\{\{state\}\}">x</figure-callout>}}^2+{\mathrm{\&sigma;}}_u^2)D}\right)& D<\frac{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="x"\; filenames="HSA00000486492200012.png"\; state="\{\{state\}\}">x</figure-callout>}}^2{\mathrm{\&sigma;}}_u^2}{({\mathrm{\&sigma;}}_x^2+{\mathrm{\&sigma;}}_u^2)}\\ 0& \mathrm{else}\end{array}\right.---\left(1\right)$

In the formula, R is the code check of Wyner-Ziv frame, and unit is a bit,

Be respectively the variance of Wyner-Ziv frame and correlated noise, D is quantizing distortion;

4.2, because

So above-mentioned rate-distortion model abbreviation is:

$R=\frac{1}{2}{\log }_2\left(\frac{{\mathrm{\&sigma;}}_u^2}{D}\right)---\left(2\right)$

4.3 the quantizing distortion D of uniform quantization is the function of quantization step Δ, the D=Δ ²/ 12, with its substitution formula (2), and consider the compression performance that different channels is encoded, obtain the BRD-SCC model:

R _band＝a·log ₂(αΔ)+b (3)

In the formula, R _BandBe defined as the check digit number and the information digit purpose ratio of coefficient tape, α, Δ are respectively Laplce's parameter and quantization step,

Be the noise variance between side information coefficient tape and the Wyner-Ziv frame coefficient tape, a, b are model coefficient, and they are the array of two dimension 1 * 16, represent 16 coefficient tapes of 4 * 4DCT.

The advantage of above-mentioned BRD-SCC model is: for different channel coding schemes, the BRD-SCC model coefficient flexibility and changeability of video sequence can be determined BRD-SCC model coefficient a and b by training process.

Realization of the present invention also is: training process, its concrete steps comprise (with coefficient tape X _iThe BRD-SCC model coefficient be example):

5.1,, calculate correlated noise model parameter α in conjunction with the DCT coefficient tape of side information and Wyner-Ziv frame _i

5.2, with two typical quantization matrix Q1 and Q2, quantize training Wyner-Ziv frame respectively, obtain coefficient tape T _iTwo quantization step Δs ' _iAnd Δ " _i, and extract bit-plane;

5.3, calculate under two kinds of quantization schemes coefficient tape X respectively _iCheck digit number and information digit purpose ratio

With

5.4, two rate distortion points that utilization obtains above, design factor band X _iBRD-SCC model coefficient a _iAnd b _i:

$\left\{\begin{array}{c}{a}_i(R_{{\mathrm{band}}_i}^{\&prime;}-R_{{\mathrm{band}}_i}^{\&prime;\&prime;})/{\log }_2({{\mathrm{\&Delta;}}^{\&prime;}}_i/{{\mathrm{\&Delta;}}^{\&prime;\&prime;}}_i)\\ b_i=R_{{\mathrm{band}}_i}^{\&prime;}-a_i\&CenterDot;{\log }_2({\mathrm{\&alpha;}}_i\&CenterDot;{{\mathrm{\&Delta;}}_i}^{\&prime;})\end{array}\right.---\left(4\right)$

5.5, repeat above-mentioned 4 steps, obtain all BRD-SCC coefficients.

By above-mentioned training process, can obtain corresponding with it BRD-SCC model coefficient at different bit-plane Data Rate Distribution schemes, realized dynamic adjustment, to estimate code check more accurately.

Realization of the present invention also is: select the quantization matrix process, its concrete steps comprise:

6.1, reading the forerunner's key frame and the follow-up key frame of current Wyner-Ziv frame, the simple motion compensated interpolation generates side information;

6.2, side information is carried out 4 * 4DCT, and extraction coefficient band Y _i, the residual error of it and Wyner-Ziv frame coefficient tape is obeyed laplacian distribution;

6.3, design factor band X _iLaplce's parameter alpha _i

6.4, corresponding to the quantization matrix of determining, design factor band X _iThe quantization step Δ _i

6.5, with the 6.3 Laplce's parameter alpha that obtain _iWith the 6.4 quantization step Δs that obtain _iIn the substitution BRD-SCC model, estimate the code check of each coefficient tape, and calculate the estimation code check of Wyner-Ziv frame;

6.6, repeat 6.3 to 6.5, calculate the estimation code check of eight quantization matrix correspondences;

6.7 comparison object code check and eight estimate code checks one by one, if the estimation code check of quantization matrix j correspondence then selects quantization matrix j to quantize current Wyner-Ziv frame near target bit rate.

After selecting the optimal quantization matrix, the Wyner-Ziv frame is quantized and the extraction bit-plane, carry out the bit-plane Data Rate Distribution, finish the Rate Control of Wyner-Ziv frame.

The present invention has made up the Wyner-Ziv frame Rate Control system based on the Wyner-Ziv video coding, and set Wyner-Ziv frame bit rate control method on this basis based on the Wyner-Ziv video coding, set up the BRD-SCC model, this is a kind of rate-distortion model with source encoding and chnnel coding combination, different by model coefficient a and b are suitable for different channel resources.The present invention determines thus that by setting up the code check that comes the estimation coefficient band in conjunction with the coefficient tape rate-distortion model of signal source and channel quantization matrix is more accurate, thereby has improved the distortion performance of DVC.

The present invention compared with prior art has following advantage:

1) the present invention has set up the BRD-SCC model, and utilize the estimation code check of the different quantization matrix correspondences of BRD-SCC Model Calculation, from alternative quantization matrix, select quantization matrix according to Wyner-Ziv frame target bit rate, this method is considered the constraint of target bit rate, has realized the purpose of reasonable use channel resource more;

2) the present invention determines BRD-SCC model coefficient a and b by training, for different bit-plane Data Rate Distribution schemes, can determine suitable BRD-SCC model coefficient by training, to guarantee the accuracy of BRD-SCC model, realize Rate Control more accurately, finally can improve the distortion performance of DVC system.

Description of drawings

Fig. 1 is a Wyner-Ziv video coding system block diagram, and wherein dotted line adds frame partly for the present invention is based on the Wyner-Ziv frame Rate Control system formation schematic diagram of Wyner-Ziv video coding;

Fig. 2 is two used typical quantization matrixes of training process of the present invention, and wherein Fig. 2 (a) is quantization matrix Q1, and Fig. 2 (b) is quantization matrix Q2;

Fig. 3 is the schematic flow sheet that the present invention is based on the Wyner-Ziv frame bit rate control method of Wyner-Ziv video coding;

Fig. 4 is Wyner-Ziv frame bit rate control method of the present invention and existing Wyner-Ziv frame bit rate control method, the distortion performance comparison diagram of hall sequence;

Fig. 5 is Wyner-Ziv frame bit rate control method of the present invention and existing Wyner-Ziv frame bit rate control method, the distortion performance comparison diagram of foreman sequence;

Fig. 6 is Wyner-Ziv frame bit rate control method of the present invention and existing Wyner-Ziv frame bit rate control method, the distortion performance comparison diagram of coastguard sequence.

Embodiment

Embodiment 1

Referring to Fig. 1, the present invention is based on the Rate Control system of the Wyner-Ziv frame of Wyner-Ziv video coding.

Fig. 1 is a Wyner-Ziv coding and decoding video whole system, the present invention proposes Wyner-Ziv frame Rate Control system based on the Wyner-Ziv video coding and is in dotted line and adds position shown in the frame, this system comprises training module, rate estimation module and choose module, training module receives first Wyner-Ziv frame of sequence to be encoded, utilize two typical quantization matrixes that it is trained, obtain the BRD-SCC model coefficient, in the rate estimation module, utilize the estimation code check of eight quantization matrix correspondences of BRD-SCC Model Calculation, in choosing module, select to satisfy the quantization matrix of target bit rate, concrete is exactly to estimate that with eight code checks compare with target bit rate, they near the time obtain estimating that code check corresponding quantitative matrix is as the optimal quantization matrix.

The input that the present invention is based on the Wyner-Ziv frame Rate Control system of Wyner-Ziv video coding is that current Wyner-Ziv frame and the distribution of frame layer bit rate obtain Wyner-Ziv frame target bit rate, output is to be the selected quantization matrix of Wyner-Ziv frame to be encoded, insert in the uniform quantizer, the function of each module is as follows:

Training module: quantize training frames respectively with two typical quantization matrix Q1 and Q2, obtain two rate distortion points, to calculate the BRD-SCC model coefficient, the BRD-SCC model of each coefficient tape is determined thereupon; Two typical quantization matrix Q1 and Q2 see Fig. 2, and be specific as follows:

$Q1=\left[\begin{array}{cccc}7& 6& 5& 4\\ 6& 5& 4& 3\\ 5& 4& 3& 3\\ 4& 3& 3& 0\end{array}\right]$ $\mathrm{<figure-callout\; id="2"\; label="Q"\; filenames="HSA00000486492200012.png"\; state="\{\{state\}\}">Q</figure-callout>}2=\left[\begin{array}{cccc}4& 3& 2& 2\\ 3& 2& 2& 2\\ 2& 2& 2& 2\\ 2& 2& 2& 0\end{array}\right]$

The rate estimation module: for Wyner-Ziv frame to be encoded, the BRD-SCC model of determining with training comes estimation coefficient tape code rate, corresponding to eight alternative quantization matrixes, estimates code check for eight that obtain the Wyner-Ziv frame;

Choose module: distribute the Wyner-Ziv frame target bit rate that obtains as input the frame layer bit rate, from eight alternative quantization matrixes, promptly estimate code check corresponding quantitative matrixes for eight, choose meet Wyner-Ziv frame target bit rate the optimal quantization matrix as output information.

In the Wyner-Ziv video coding system, the present invention selects the optimal quantization matrix for the Wyner-Ziv frame and quantizes the Wyner-Ziv frame, carries out the bit-plane Data Rate Distribution, makes that the Rate Control of Wyner-Ziv frame is more accurate.

Embodiment 2

The present invention still is a kind of Wyner-Ziv frame bit rate control method based on the Wyner-Ziv video coding, and the application of this method is to move in above-mentioned Wyner-Ziv frame Rate Control system based on the Wyner-Ziv video coding, and with reference to Fig. 3, performing step includes:

Step 1, the Wyner-Ziv frame that reads first GOP are carried out 4 * 4DCT and extraction coefficient band T as training frames to it _i(i=1,2 ..., 16);

Step 2, training utilizes two typical quantization matrix Q1 and Q2 to quantize the coefficient tape of training frames respectively, generates two rate distortion points, calculates BRD-SCC model coefficient a and b.

Step 3 reads Wyner-Ziv frame to be encoded, and it is carried out 4 * 4DCT, and extraction coefficient band X _i

Step 4, select quantization matrix, the BRD-SCC model coefficient a and the b that utilize BRD-SCC model and training to obtain, calculate the estimation code check of eight quantization matrix correspondences, estimate that with eight code checks compare with the Wyner-Ziv frame target bit rate that the control of frame layer bit rate obtains, select and the pairing quantization matrix of the immediate estimation code check of target bit rate.The present invention chooses the optimal quantization matrix from eight alternative quantization matrixes.

Step 5 is utilized the quantization matrix of selecting to obtain to quantize the Wyner-Ziv frame, and is extracted bit-plane.

Step 6 is carried out the bit-plane Data Rate Distribution according to the correlation of side information and source information.

Step 7, step 3 is to 6 being the Rate Control process of a Wyner-Ziv frame, repeating step 3 to 6 can be finished the Rate Control of all Wyner-Ziv frames of whole video sequence.

BRD-SCC model of the present invention to set up process as follows:

4.1, the pixel domain rate-distortion model of Wyner-Ziv frame:

$R=\left\{\begin{array}{cc}\frac{1}{2}{\log }_2\left(\frac{{\mathrm{\&sigma;}}_x^2{\mathrm{\&sigma;}}_u^2}{({\mathrm{\&sigma;}}_x^2+{\mathrm{\&sigma;}}_u^2)D}\right)& D<\frac{{\mathrm{\&sigma;}}_x^2{\mathrm{\&sigma;}}_u^2}{({\mathrm{\&sigma;}}_x^2+{\mathrm{\&sigma;}}_u^2)}\\ 0& \mathrm{else}\end{array}\right.---\left(5\right)$

In the formula, R is the code check of Wyner-Ziv frame, and unit is a bit, Be respectively the variance of Wyner-Ziv frame and correlated noise, D is quantizing distortion;

4.2, because

So above-mentioned rate-distortion model abbreviation is:

$R=\frac{1}{2}{\log }_2\left(\frac{{\mathrm{\&sigma;}}_u^2}{D}\right)---\left(6\right)$

4.3 the quantizing distortion D of uniform quantization is the function of quantization step Δ, the D=Δ ²/ 12, with its substitution formula (2), and consider the compression performance that different channels is encoded, obtain the BRD-SCC model:

R _band＝a·log ₂(αΔ)+b (7)

In the formula, R _BandBe defined as the check digit number and the information digit purpose ratio of coefficient tape, α, Δ are respectively Laplce's parameter and quantization step,

Be the noise variance between side information coefficient tape and the Wyner-Ziv frame coefficient tape, a, b are model coefficient, and they are the array of two dimension 1 * 16, represent 16 coefficient tapes of 4 * 4DCT.

Wherein, the training process of step 2, concrete steps are as follows:

5.1, reading the forerunner's key frame and the follow-up key frame of training frames, the simple motion compensated interpolation generates side information, and side information is carried out 4 * 4DCT, extraction coefficient band Y _i, and combined training frame coefficient tape T _iThe noise variance that calculates the coefficient of correspondence band is:

${\mathrm{\&sigma;}}_{u_i}^2=E\left[{(T_i-Y_i)}^2\right]-{\left(E\right[T_i-Y_i\left]\right)}^2---\left(8\right)$

In the formula, E[] be to ask the expectation computing;

So, Laplce's parameter alpha of the correlated noise model of i coefficient tape _iFor:

${\mathrm{\&alpha;}}_i=\sqrt{\frac{2}{{\mathrm{\&sigma;}}_{u_i}^2}}---\left(9\right)$

5.2, quantize training frames coefficient tape T respectively with two typical quantization matrix Q1 and Q2 _i, under two kinds of quantization schemes, obtain two quantization step Δs ' _i, Δ " _i, extract bit-plane;

If M _{I, j}Coefficient tape X when expression quantizes the Wyner-Ziv frame with quantization matrix j (j=1,2..., 8) _iThe corresponding quantitative bit number is so the quantizing bit number of Wyner-Ziv frame is the bit number sum of all coefficient tapes, that is:

$R_{s_j}=\underset{i=1}{\overset{16}{\mathrm{\&Sigma;}}}L\&CenterDot;M_{i,j}---\left(10\right)$

In the formula, L is the length of coefficient tape, is that the image of QCIF (176 * 144) carries out 4 * 4DCT to normal size for example, and the length that obtains the luminance factor band is L=176 * 144/ (4 * 4)=1584;

5.3, according to certain bit-plane Data Rate Distribution scheme, calculate the code check of each bit-plane, and calculate the average bit rate of whole coefficient tape Coefficient tape X _iThe code check computing formula be:

$\left\{\begin{array}{c}{R}_{{\mathrm{band}}_i}^{\&prime;}=\frac{1}{M_i^{\&prime;}}\underset{b=1}{\overset{M_i^{\&prime;}}{\mathrm{\&Sigma;}}}{R}_b^{\&prime;}\\ R_{{\mathrm{band}}_i}^{\&prime;\&prime;}=\frac{1}{M_i^{\&prime;\&prime;}}\underset{b=1}{\overset{M_i^{\&prime;\&prime;}}{\mathrm{\&Sigma;}}}{R}_b^{\&prime;\&prime;}\end{array}\right.---\left(11\right)$

In the formula, R ' _b, R " _bBe respectively b the bit-plane check bit number of typical quantization matrix Q1 and Q2 correspondence and the ratio of information bit.M ' _i, M " _iBe respectively typical quantization matrix Q1 and Q2 to coefficient tape T _iQuantizing bit number;

5.4, utilize step 1 to 4 two rate distortion points that generate

With

Calculate coefficient tape T _iBRD-SCC model coefficient a _iAnd b _i:

$\left\{\begin{array}{c}{a}_i(R_{{\mathrm{band}}_i}^{\&prime;}-R_{{\mathrm{band}}_i}^{\&prime;\&prime;})/{\log }_2({{\mathrm{\&Delta;}}^{\&prime;}}_i/{{\mathrm{\&Delta;}}^{\&prime;\&prime;}}_i)\\ b_i=R_{{\mathrm{band}}_i}^{\&prime;}-a_i\&CenterDot;{\log }_2({\mathrm{\&alpha;}}_i\&CenterDot;{{\mathrm{\&Delta;}}_i}^{\&prime;})\end{array}\right.---\left(12\right)$

5.5, repeat above-mentioned 4 steps, obtain all BRD-SCC coefficients.

Step 4 of the present invention is selected the quantization matrix process, and concrete steps are as follows:

6.1, reading the forerunner's key frame and the follow-up key frame of Wyner-Ziv frame to be encoded, the simple motion compensated interpolation generates side information;

6.2, side information is carried out 4 * 4DCT, extraction coefficient band Y _i, Y _iWith Wyner-Ziv frame coefficient tape X _iResidual error obey laplacian distribution;

6.3, calculate Wyner-Ziv frame coefficient tape X _iThe laplacian distribution parameter alpha _i

${\mathrm{\&alpha;}}_i=\sqrt{\frac{2}{{\mathrm{\&sigma;}}_{u_i}^2}}---\left(13\right)$

In the formula,

Be coefficient tape X _iThe correlated noise variance, it be calculated as follows:

${\mathrm{\&sigma;}}_{u_i}^2=E{[X_i-Y_i]}^2-{\left(E\right[X_i-Y_i\left]\right)}^2---\left(14\right)$

6.4, corresponding to the quantization matrix of determining, design factor band X _iThe quantization step Δ _iQuantization step is calculated and can be got by following formula:

${\mathrm{\&Delta;}}_{i,j}=\left\{\begin{array}{cc}({\mathrm{mc}}_i-2^{M_{i,j}}-1)/2^{M_{i,j}}& i=1\\ (2\&CenterDot;{\mathrm{mc}}_i-2^{M_{i,j}}-2)/(2^{M_{i,j}}-1)& i>1\end{array}\right.---\left(15\right)$

Wherein, M _{I, j}When quantizing the Wyner-Ziv frame for usefulness quantization matrix j, coefficient tape X _iThe bit-plane number; Mc _iBe coefficient tape X _iThe maximum value of interior coefficient, the quantization step of the corresponding DC coefficient tape of i=1, the quantization step of i＞1 corresponding A C coefficient tape;

6.5, with the 6.3 Laplce's parameter alpha that obtain _iWith the 6.4 quantization step Δs that obtain _iIn the substitution BRD-SCC model, estimate the code check of each coefficient tape, and calculate the estimation code check of Wyner-Ziv frame;

6.6, repeat 6.3 to 6.5, calculate the estimation code check of eight quantization matrix correspondences;

6.7, estimate code check with eight

Compare with target bit rate one by one, if quantization matrix j ^*Corresponding estimation code check

The most approaching with target bit rate, then select quantization matrix j ^*Quantize the DCT coefficient tape of current Wyner-Ziv frame.

Embodiment 3

Based on the Wyner-Ziv frame Rate Control system of Wyner-Ziv video coding with embodiment 1, based on the Wyner-Ziv frame bit rate control method of Wyner-Ziv video coding with embodiment 2.This example is specifically calculated selecting quantization matrix, and wherein step 6.1-6.4 in the step 6.5, utilizes BRD-SCC model and the definite model coefficient estimation coefficient tape code rate of training with embodiment 2, when then quantizing the Wyner-Ziv frame with quantization matrix j, and coefficient tape X _iCode check be:

$R_{{\mathrm{band}}_{i,j}}=a_i\&CenterDot;{\log }_2({\mathrm{\&alpha;}}_i\&CenterDot;{\mathrm{\&Delta;}}_{i,j})+b_i---\left(16\right)$

In quantization step formula (15) substitution BRD-SCC modular form (16), obtain the coefficient tape X of quantization matrix j correspondence _iThe estimation code check:

$R_{i,j}=\left\{\begin{array}{cc}{a}_i\&CenterDot;{\log }_2\left(\frac{{\mathrm{\&alpha;}}_i({\mathrm{mc}}_i-2^{M_{i,j}}-1)}{2^{M_{i,j}}}\right)+b_i& i=1\\ a_i\&CenterDot;{\log }_2\left(\frac{{\mathrm{\&alpha;}}_i(2\&CenterDot;{\mathrm{mc}}_i-2^{M_{i,j}}-2)}{2^{M_{i,j}}-1}\right)+b_i& i>1\end{array}\right.---\left(17\right)$

When utilizing formula (17) can calculate each quantization matrix quantification Wyner-Ziv frame, the estimation code check of 16 coefficient tapes of 4 * 4DCT;

Estimate the Wyner-Ziv frame estimation code check of quantization matrix j correspondence

The total bitrate that the code check of 16 coefficient tapes of Wyner-Ziv frame is added and obtain the Wyner-Ziv frame is:

${\hat{R}}_J=\underset{i=1}{\overset{16}{\mathrm{\&Sigma;}}}{R}_{i,j}\&CenterDot;L\&CenterDot;M_{i,j}---\left(18\right)$

In the formula, L is the length of coefficient tape, can calculate the estimation code check of eight alternative quantization matrix correspondences with formula (18), when j is 1, and the estimation code check of corresponding first quantization matrix When j is 2, the estimation code check of corresponding second quantization matrix

When j is 8, the estimation code check of corresponding the 8th quantization matrix Step 6.6-6.7 afterwards is with embodiment 2.

Embodiment 4

Based on the Wyner-Ziv frame Rate Control system of Wyner-Ziv video coding with embodiment 1, based on the Wyner-Ziv frame bit rate control method of Wyner-Ziv video coding with embodiment 2-3.

In order to clearly demonstrate the present invention and the related of Wyner-Ziv video coding and effect, at first the coding side performing step to the Wyner-Ziv video coding system is summarized as follows:

One, input video sequence is divided into key frame and Wyner-Ziv frame, key frame and Wyner-Ziv frame composition length are 2 GOP;

Two,, adopt H.264/AVC intraframe coding for key frame;

Three, the Wyner-Ziv frame is carried out 4 * 4DCT, and the extraction coefficient band;

Four, Wyner-Ziv frame coefficient tape is quantized, and extract bit-plane;

Five, bit-plane is carried out the LDPC coding, generate check digit, be transferred to the auxiliary LDPC decoding of decoding end.

After the Wyner-Ziv video coding system carries out step 2 extraction coefficient band, quantize by empirical value in the prior art with key frame quantization parameter coupling, and the present invention need calculate the estimation code check of each quantization matrix correspondence at this, the Wyner-Ziv frame target bit rate that the control of they and frame layer bit rate is obtained compares, choose suitable quantization matrix, for the quantizing process of step 4 is prepared, used quantization matrix is a quantization matrix provided by the invention when whole Wyner-Ziv video coding system carries out step 4.

Embodiment 5

Based on the Wyner-Ziv frame Rate Control system of Wyner-Ziv video coding with embodiment 1, based on the Wyner-Ziv frame bit rate control method of Wyner-Ziv video coding with embodiment 2-4.

Respectively the present invention and existing method are used for the Wyner-Ziv video coding and decoding system, to frame per second is that the video sequence Hall of 15Hz carries out encoding and decoding, with Y-PSNR (Peak Signal-Noise Ratio, PSNR) objective quality of measurement decoding and rebuilding video, relatively under same code rate, system's distortion performance of two kinds of Wyner-Ziv frame bit rate control methods, experiment condition is as follows:

Hardware environment: CPUAMD Sempron 3000+, 1.8GHz, 512MB internal memory;

Software environment: Windows XP, Microsoft Visual Studio.NET 2003;

Image size: QCIF (176 * 144);

GOP length: 2;

Sequence length: 300 frames.

Experimental result as can be seen from the figure, for the hall sequence, under same code rate, adopts the bit rate control method of the present invention's proposition to improve about 1dB than the existing bit rate control method PSNR of employing as shown in Figure 4.

Embodiment 6

Based on the Wyner-Ziv frame Rate Control system of Wyner-Ziv video coding with embodiment 1, based on the Wyner-Ziv frame bit rate control method of Wyner-Ziv video coding with embodiment 2-4.

Respectively the present invention and existing method are used for Wyner-Ziv video coding system as embodiment 1, to frame per second is that the video sequence Foreman of 30Hz carries out encoding and decoding, experiment condition is with embodiment 5, weigh the objective quality of decoding and rebuilding video with PSNR, under same code rate, compare system's distortion performance of two kinds of Wyner-Ziv frame bit rate control methods.

Experimental result such as Fig. 5, as can be seen from the figure, for the Foreman sequence, under same code rate, the present invention has improved about 0.5dB than existing bit rate control method PSNR.

Embodiment 7

Based on the Wyner-Ziv frame Rate Control system of Wyner-Ziv video coding with embodiment 1, based on the Wyner-Ziv frame bit rate control method of Wyner-Ziv video coding with embodiment 2-4.

Respectively the present invention and existing method are applied to the Wyner-Ziv video coding system, to frame per second is that the video sequence Coastguard of 30Hz carries out encoding and decoding, experiment condition is with embodiment 5, weigh the objective quality of decoding and rebuilding video with PSNR, under same code rate, compare system's distortion performance of two kinds of Wyner-Ziv frame bit rate control methods.

Experimental result such as Fig. 6, as can be seen from the figure, for the Coastguard sequence, under same code rate, the present invention has improved about 0.5dB than existing bit rate control method PSNR.

By above experimental result as can be known: the Wyner-Ziv frame bit rate control method that adopts the present invention to propose, can raising in various degree there be the distortion performance of feeding back the distributed video coding system, for the less video sequence of amount of exercise, because determining of BRD-SCC model coefficient is more accurate, the distortion performance raising is more; Otherwise, the video sequence that amount of exercise is bigger, distortion performance improves less.

The present invention determines thus that by setting up the code check that comes the estimation coefficient band in conjunction with the coefficient tape rate-distortion model of signal source and channel quantization matrix is more accurate, thereby has improved the distortion performance of DVC.Code rate allocation method of the present invention can be used for not having in the Wyner-Ziv video coding system of feedback Rate Control.

Claims (6)

1. Wyner-Ziv frame Rate Control system based on the Wyner-Ziv video coding, it is characterized in that: this system includes: training module, the rate estimation module, choose module, training module receives first Wyner-Ziv frame of sequence to be encoded, utilize two typical quantization matrixes to train, obtain coefficient tape rate-distortion model coefficient in conjunction with signal source and channel, in the rate estimation module, utilize estimation code check in conjunction with eight quantization matrix correspondences of coefficient tape rate distortion BRD-SCC Model Calculation of signal source and channel, module is chosen in access, in choosing module, eight estimation code checks are compared with target bit rate, select to satisfy the quantization matrix of target bit rate.

2. the Wyner-Ziv frame Rate Control system of the described Wyner-Ziv video coding of claim 1, it is characterized in that: described two typical quantization matrix Q1 and Q2 are respectively:

$Q1=\left[\begin{array}{cccc}7& 6& 5& 4\\ 6& 5& 4& 3\\ 5& 4& 3& 3\\ 4& 3& 3& 0\end{array}\right]$ $Q2=\left[\begin{array}{cccc}4& 3& 2& 2\\ 3& 2& 2& 2\\ 2& 2& 2& 2\\ 2& 2& 2& 0\end{array}\right]$

3. Wyner-Ziv frame bit rate control method based on the Wyner-Ziv video coding, it is characterized in that: according to the selection quantization matrix of encoder bit rate change dynamics, the step of this control method includes:

Step 1, the Wyner-Ziv frame of choosing first GOP are carried out 4 * 4 discrete cosine transform DCT as training frames to it, and the extraction coefficient band;

Step 2, training utilizes two typical quantization matrix Q1 and Q2 to quantize the coefficient tape of training frames respectively, generates two rate distortion points, calculates BRD-SCC model coefficient a and b;

Step 3 reads current Wyner-Ziv frame, and it is carried out 4 * 4DCT, and the extraction coefficient band;

Step 4, select quantization matrix, BRD-SCC model coefficient a and b that usage factor band rate-distortion model and training obtain, calculate the estimation code check of eight quantization matrix correspondences, estimate that with eight code checks compare with the Wyner-Ziv frame target bit rate that the control of frame layer bit rate obtains, select and the pairing quantization matrix of the immediate estimation code check of target bit rate;

Step 5 is utilized the quantization matrix of selecting to obtain to quantize the Wyner-Ziv frame, and is extracted bit-plane;

Step 6 is carried out the bit-plane Data Rate Distribution according to the correlation of side information and source information;

Step 7, step 3 is to 6 being the Rate Control process of a Wyner-Ziv frame, repeating step 3 to 6 can be finished the Rate Control of all Wyner-Ziv frames of whole video sequence.

4. the described Wyner-Ziv frame bit rate control method based on the Wyner-Ziv video coding of claim 3 is characterized in that: described BRD-SCC model to set up process as follows:

4.1, the pixel domain rate-distortion model of Wyner-Ziv frame:

$R=\left\{\begin{array}{cc}\frac{1}{2}{\log }_2\left(\frac{{\mathrm{\&sigma;}}_x^2{\mathrm{\&sigma;}}_u^2}{({\mathrm{\&sigma;}}_x^2+{\mathrm{\&sigma;}}_u^2)D}\right)& D<\frac{{\mathrm{\&sigma;}}_x^2{\mathrm{\&sigma;}}_u^2}{({\mathrm{\&sigma;}}_x^2+{\mathrm{\&sigma;}}_u^2)}\\ 0& \mathrm{else}\end{array}\right.$

In the formula, R is the code check of Wyner-Ziv frame, and unit is a bit, Be respectively the variance of Wyner-Ziv frame and correlated noise, D is quantizing distortion;

4.2, because

So above-mentioned rate-distortion model abbreviation is:

$R=\frac{1}{2}{\log }_2\left(\frac{{\mathrm{\&sigma;}}_u^2}{D}\right)$

4.3 the quantizing distortion D of uniform quantization is the function of quantization step Δ, the D=Δ ²/ 12, with its substitution following formula, and consider the compression performance that different channels is encoded, obtain the BRD-SCC model:

R _band＝a·log ₂(αΔ)+b

In the formula, R _BandBe defined as the check digit number and the information digit purpose ratio of coefficient tape, α, Δ are respectively Laplce's parameter and quantization step,

Be the noise variance between side information coefficient tape and the Wyner-Ziv frame coefficient tape, a, b are model coefficient, and they are the array of two dimension 1 * 16, represent 16 coefficient tapes of 4 * 4DCT.

5. the described Wyner-Ziv frame bit rate control method based on the Wyner-Ziv video coding of claim 3 is characterized in that: training process, and concrete steps are as follows:

5.1,, calculate correlated noise model parameter α in conjunction with the DCT coefficient tape of side information and Wyner-Ziv frame _i

5.2, with two typical quantization matrix Q1 and Q2, quantize training Wyner-Ziv frame respectively, obtain coefficient tape T _iTwo quantization step Δs ' _iAnd Δ " _i, and extract bit-plane;

5.3, calculate under two kinds of quantization schemes coefficient tape X respectively _iCheck digit number and information digit purpose ratio

With

5.4, two rate distortion points that utilization obtains above, design factor band X _iBRD-SCC model coefficient a _iAnd b _i:

$\left\{\begin{array}{c}{a}_i(R_{{\mathrm{band}}_i}^{\&prime;}-R_{{\mathrm{band}}_i}^{\&prime;\&prime;})/{\log }_2({{\mathrm{\&Delta;}}^{\&prime;}}_i/{{\mathrm{\&Delta;}}^{\&prime;\&prime;}}_i)\\ b_i=R_{{\mathrm{band}}_i}^{\&prime;}-a_i\&CenterDot;{\log }_2({\mathrm{\&alpha;}}_i\&CenterDot;{{\mathrm{\&Delta;}}_i}^{\&prime;})\end{array}\right.$

5.5, repeat above-mentioned 4 steps, obtain all BRD-SCC coefficients.

6. the described Wyner-Ziv frame bit rate control method based on the Wyner-Ziv video coding of claim 3 is characterized in that selecting the quantization matrix process, and concrete steps are as follows:

6.1, reading the forerunner's key frame and the follow-up key frame of current Wyner-Ziv frame, the simple motion compensated interpolation generates side information;

6.2, side information is carried out 4 * 4DCT, and extraction coefficient band Y _i, the residual error of it and Wyner-Ziv frame coefficient tape is obeyed laplacian distribution;

6.3, design factor band X _iLaplce's parameter alpha _i

6.4, corresponding to the quantization matrix of determining, design factor band X _iThe quantization step Δ _i

6.5, with the 6.3 Laplce's parameter alpha that obtain _iWith the 6.4 quantization step Δs that obtain _iIn the substitution BRD-SCC model, estimate the code check of each coefficient tape, and calculate the estimation code check of Wyner-Ziv frame;

6.6, repeat 6.3 to 6.5, calculate the estimation code check of eight quantization matrix correspondences;

6.7 comparison object code check and eight estimate code checks one by one, if the estimation code check of quantization matrix j correspondence then selects quantization matrix j to quantize current Wyner-Ziv frame near target bit rate.

Images