CN102045571A - Fast iterative search algorithm for stereo video coding - Google Patents

Abstract

The invention discloses a fast iterative search algorithm for stereo video coding, which is characterized in that a stereo-motion constraint model is defined according to the relation between the motion vector and disparity vector of the stereo image pair of the adjacent images of the left viewpoint and right viewpoint of a stereo video. The fast iterative search method for stereo video coding comprises the following steps: initializing, adjusting the RSR (rank sum ration) of an amending search window, carrying out iterative search, and suspending: calculating the initial motion vector and initial disparity vector of a current module by adopting an iterative search strategy, and designing the adaptive amending search window according to the error of the stereo-motion constraint model to amend the motion vector and disparity vector of the current module so that the optimum motion vector and optimum disparity vector of the current module can be finally predicted rapidly. Compared with the traditional full search algorithm, the fast iterative search algorithm provided by the invention can ensure the coding quality and save the coding time by over 96 percent.

Description

A kind of stereo scopic video coding iteratively faster searching method

Technical field

The present invention relates to field of video encoding, especially relate to motion vector and difference vector fast search algorithm in a kind of stereo scopic video coding.

Background technology

Three-dimensional video-frequency contains the depth information of scenery, has more the sense of reality on the sign of natural scene, has represented wide application prospect at the stereo visual system of 3D TV, mobile device and fields such as video conferencing with telepresenc.

Two video channels about three-dimensional video-frequency comprises, typical IPPPP predict as shown in Figure 1, horizontal direction is a time orientation, vertical direction is a viewpoint direction.Make that left viewpoint is a reference view, promptly left viewpoint is coding earlier, and first frame of left viewpoint is the I frame, when coding, does not need the information with reference to other frame, directly carries out dct transform, equal interval quantizing, and run length coding is sent into arithmetic encoder at last.Viewpoint other frame except that first frame in a left side all is the P frame, carries out estimation by the coded frame with reference to the left previous moment of viewpoint.Right viewpoint is the prediction viewpoint, and first frame is the P frame, has both allowed it to carry out disparity estimation with reference to first frame of left viewpoint, allows intraframe predictive coding again, chooses more excellent coded system from the two, has guaranteed code efficiency.All the other P frames of right viewpoint all comprise two reference frames, the reference frame of not only wanting the reference time direction (promptly, the coded frame in the right previous moment of viewpoint) carries out estimation, also want the reference frame (that is the coded frame in the left identical moment of viewpoint) of reference view direction to carry out disparity estimation.Traditional three-dimensional video-frequency compression passing through all direction search method, adopt the large search window to carry out estimation and disparity estimation respectively, with the time and space redundancy of eliminating same viewpoint inside and the crossing redundancy between the viewpoint of the left and right sides, and the rate distortion costs of comparing motion vector and difference vector, selection makes the final predictive vector as current block of rate distortion costs minimum.Wherein, rate distortion costs by RDCost (mv)=SAD (c, r)+λ * R (mv-p) calculates, mv represents the motion/disparity vector of current block, c represents current block, r represents to predict piece, and λ represents Lagrange multiplier, and p represents the predicted value of the motion/disparity vector of current block, the required bit number of difference of R (mv-p) presentation code motion/disparity vector and predicted value, SAD (c, r) expression current block and the prediction piece absolute error and

B ₁, B ₂Level and the Vertical number of pixels of representing piece respectively, the coordinate of [i, j] remarked pixel, c[i, j] expression current block pixel value; R[i-mv _x, j-mv _y] pixel value of expression prediction piece, (mv _x, mv _y) level and the vertical component size of motion/disparity vector of expression current block.Traditional full-search algorithm has brought huge operand when obtaining high distortion performance, limited the real-time application of three-dimensional video-frequency.

At present, three-dimensional video-frequency fast coding algorithm can be divided into two big classes substantially: a class is based on the encryption algorithm of predictive vector, with full-search algorithm a certain territory (disparity domain or motion field) calculated parallax or motion vector earlier, utilize " stereo-picture to ", to adopt fast algorithm to predict [1-2] then to another territory (motion field or disparity domain) at the consistency or the adjacent viewpoint motion vector agreement principle of adjacent moment difference vector.This class algorithm can obtain coding efficiency preferably, but because the accuracy of next territory (motion field or disparity domain) predictive vector depends on predicting the outcome of last territory (disparity domain or motion field), therefore last territory often adopts the full-search algorithm of limit to guarantee result's accuracy, and coding rate still has much room for improvement.Another kind of is motion and parallax associating algorithm for estimating, serial correlation principle according to three-dimensional video-frequency, the information of motion field and disparity domain can be utilized mutually, motion and the direct prediction of difference vector relation by adjacent image obtain the motion/disparity vector of current block, thereby reduce encoder complexity [3-4] to greatest extent.But the research of this class algorithm at present is mostly only at pixel domain or based on mpeg standard, can not with the block-based H.264/AVC video encoding standard compatibility of current main-stream, and, directly utilize the motion and the difference vector of adjacent image to concern that trying to achieve predictive vector is absorbed in local minimum easily, coding quality can not get guaranteeing.

The present invention is based on H.264/AVC standard, propose a kind of stereo scopic video coding iteratively faster searching algorithm based on solid-kinematic constraint model, significantly reduce encoder complexity under the prerequisite that guarantees high compression rate, is very significant.

Attached: list of references

[1]Ding?L?F，Chien?S?Y，Chen?L?G.Joint?prediction?algorithm?and?architecture?for?stereo?video?hybrid?coding?systems[J]IEEE?Transactions?on?Circuits?and?Systems?for?Video?Technology，2006，16(11)：1324-1337

[2]Lai?P，Ortega?A.Predictive?fast?motion/disparity?search?for?multiview?video?coding[C]//SPIE.Proceedings?of?SPIE.San?Jose：Visual?Communications?and?Image?Processing，2006，6077：607709

[3]Paras?I，Alvertos?N，Tziritas?G.Joint?disparity?and?motion?field?estimation?in?stereoscopic?image?sequences[C]//IEEE.Proceedings?of?13th?International?Conference?on?Pattern?Recognition.Vienna：ICPR，1996：359-363

[4]Kim?Y，Lee?J，Park?C，et?al.MPEG-4?compatible?stereoscopic?sequence?codec?for?stereo?broadcasting[J]IEEE?Transactions?on?Consumer?Electronics，2005，51(4)：1227-1236

Summary of the invention

The objective of the invention is to,,, realize the stereo scopic video coding of low complex degree to solve the high problem of picture frame encoder complexity of the right viewpoint of three-dimensional video-frequency except that first frame by a kind of stereo scopic video coding iteratively faster searching method is provided.

The present invention solves the problems of the technologies described above the technical scheme of taking:

A kind of stereo scopic video coding iteratively faster searching method makes the macro block MB in the right viewpoint t time chart picture _{R, t}Be current block, i is the iteration step number, and δ is the model error of solid-kinematic constraint model, mv _{R, t}(MB _{R, t}) expression current block the optimal motion vector, dv _{R, t}(MB _{R, t}) expression current block optimum difference vector,

Be the rate distortion costs of the motion vector of the i time revised current block of iteration,

Be the rate distortion costs of the motion vector of the i-1 time revised current block of iteration,

Be the rate distortion costs of the difference vector of the i time revised current block of iteration,

Be the rate distortion costs of the difference vector of the i-1 time revised current block of iteration, may further comprise the steps:

1.1, initialization; Determine the motion-vector search starting point of current block

Difference vector search starting point with current block Obtain the motion-vector search starting point of revised current block

Difference vector search starting point with revised current block

Preserve the motion of revised current block and the rate distortion costs of difference vector search starting point;

1.2, according to the size of following formula adjustment correction search window RSR,

$\mathrm{RSR}=\left\{\begin{array}{cc}{\mathrm{RSR}}_{\mathrm{MIN}}& \mathrm{\&delta;}<{\mathrm{<figure-callout\; id="1"\; label="T"\; filenames="HDA0000043764300000021.png,HDA0000043764300000041.png"\; state="\{\{state\}\}">T</figure-callout>}}_1\\ {\mathrm{RSR}}_{\mathrm{MIN}}+\frac{\mathrm{\&delta;}-T_1}{T_2-T_1}({\mathrm{RSR}}_{\mathrm{MAX}}-{\mathrm{RSR}}_{\mathrm{MIN}})& {\mathrm{<figure-callout\; id="1"\; label="T"\; filenames="HDA0000043764300000021.png,HDA0000043764300000041.png"\; state="\{\{state\}\}">T</figure-callout>}}_1\&le;\mathrm{\&delta;}\&le;{\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA0000043764300000021.png,HDA0000043764300000041.png"\; state="\{\{state\}\}">T</figure-callout>}}_2\\ {\mathrm{RSR}}_{\mathrm{MAX}}& \mathrm{\&delta;}>{\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA0000043764300000021.png,HDA0000043764300000041.png"\; state="\{\{state\}\}">T</figure-callout>}}_2\end{array}\right.$

Wherein, T ₁And T ₂Represent two threshold value (T ₁＜T ₂), RSR _MINBe minimum correction search window, RSR _MAXBe the maximum modified search window;

1.3, iterative search procedures; Determine the difference vector prediction initial value of the i time iteration current block

Carry out the vector correction, obtain the difference vector of the i time revised current block of iteration Determine the motion-vector prediction initial value of the i time iteration current block

Carry out the vector correction, obtain the motion vector of the i time revised current block of iteration

Preserve the rate distortion costs of the motion vector and the difference vector of the i time revised current block of iteration.

1.4, abort criterion: if

And

Then make the motion vector of the i-1 time revised current block of iteration

And difference vector

Respectively as the optimal motion vector mv of current block _{R, t}(MB _{R, t}) and the optimum difference vector dv of current block _{R, t}(MB _{R, t}), the finishing iteration search procedure, otherwise, make i=i+1, by

Recomputate and upgrade δ, jump to step 1.2, wherein,

With

Represent the motion vector of the i time revised current block of iteration and the difference vector of the i time revised current block of iteration respectively;

Wherein, The difference vector of representing the motion compensation block of the i time iteration current block in the time orientation reference frame,

The motion vector of the parallax compensation piece of the i time iteration current block in the viewpoint direction reference frame.

Aforesaid step 1.1 comprises:

2.1, make i=0, δ=0;

2.2, the motion-vector search starting point of described definite current block

Difference vector search starting point with current block

Pass through candidate set

With

Obtain;

Wherein, mv _a/ dv _a, mv _b/ dv _bAnd mv _c/ dv _cRepresent left side piece a, upper block b that current block is adjacent and motion or the difference vector of upper right c respectively, mv _MedAnd dv _MedRepresent the median vector of current block motion vector and the median vector of current block difference vector respectively, mv _{L, t}Be the motion vector of current block piece identical in the viewpoint direction reference frame, dv with the current block position _{R, t-1}Difference vector for current block piece identical in the time orientation reference frame with the current block position;

2.3, the described motion-vector search starting point that obtains revised current block

Difference vector search starting point with revised current block

Respectively with the motion-vector search starting point of current block Difference vector search starting point with current block

Be the center, delimit a RSR _MIN* RSR _MINThe correction search window, in this search window, do the vector correction and obtain; The rate distortion costs of the motion-vector search starting point of the revised current block of described preservation, note is done

The rate distortion costs of the difference vector search starting point of the revised current block of described preservation, note is done Make i=i+1.

Threshold value T in the aforesaid step 1.2 ₁Be 5, threshold value T ₂Be 20, RSR _MINBe 2, RSR _MAXBe 96.

Aforesaid step 1.3 comprises:

4.1, the difference vector of the i time iteration current block prediction initial value

By

Calculate;

Wherein,

The motion vector of representing the current block that the i-1 time iteration obtains; The difference vector of the encoding block that expression is covered by the motion compensation block of current block in the time orientation reference frame, u are represented the number of the encoding block that is capped, will make

Minimum

As the difference vector of the motion compensation block of the i time iteration current block in the time orientation reference frame, note is done

It is the motion vector of the parallax compensation piece of current block in the viewpoint direction reference frame that obtain of the i-1 time iteration;

With

Be the center, delimit the search window of a RSR * RSR, in this search window, carry out the vector correction, obtain the difference vector of the i time revised current block of iteration

Preserve the rate distortion costs of the difference vector of the i time revised current block of iteration, note is done

4.2, the motion-vector prediction initial value of the i time iteration current block

By

Calculate;

Wherein,

The difference vector of the i time revised current block of iteration that obtains in the expression step 4.1,

The difference vector of the encoding block that expression is covered by the parallax compensation piece of current block in the viewpoint direction reference frame, v are represented the number of the encoding block that is capped, will make

Minimum

As the motion vector of the parallax compensation piece of the i time iteration current block in the viewpoint direction reference frame, note is done

The difference vector of the motion compensation block of the i time iteration current block in the time orientation reference frame that obtains in the expression step 4.1;

With

Be the center, delimit the search window of a RSR * RSR, in this search window, carry out the vector correction, obtain the motion vector of the i time revised current block of iteration

Preserve the rate distortion costs of the motion vector of the i time revised current block of iteration, note is done

Compared with prior art, the invention has the advantages that: traditional stereo scopic video coding algorithm adopts large search window separate searches motion vector and difference vector respectively in time orientation reference frame and viewpoint direction reference frame, there is not to utilize the vector correlation between " stereo-picture to ", the present invention is by setting up solid-kinematic constraint model, and iterative search strategy according to a kind of self adaptation correction of big gizmo window of model error, this method can be when keeping coding quality, greatly reduce the complexity of stereo scopic video coding, improve coding rate.

Experimental result proves that method of the present invention can not save for average 96.43% scramble time when not reducing coding quality substantially.

Description of drawings

Fig. 1 is the stereo scopic video coding structural representation;

Fig. 2 is solid-kinematic constraint model schematic diagram;

Fig. 3 is the flow chart of the inventive method;

Fig. 4 is the predictive vector schematic diagram of adjacent block;

Fig. 5 is a schematic diagram of asking the motion-vector prediction initial value;

Fig. 6 is a schematic diagram of asking difference vector prediction initial value;

Fig. 7 is the rate distortion curve signal of " Ballroom " sequence algorithms of different.

Embodiment

Embodiment is further elaborated the present invention below in conjunction with accompanying drawing.

Fig. 2 is solid-kinematic constraint model schematic diagram, and 4 width of cloth images of three-dimensional video-frequency left and right sides viewpoint adjacent moment are called one " stereo-picture to ", and wherein, right viewpoint t time chart is as F _{R, t}Be present frame, MB _{R, t}The expression present frame, F _{R, t}In current block.F _{L, t}Expression current block MB _{R, t}At the reference frame of viewpoint direction, F _{R, t-1}Expression current block MB _{R, t}At the reference frame of time orientation, F _{L, t-1}Expression DCMB _{L, t}At the reference frame of time orientation, F _{L, t-1}Also be MCMB _{R, t-1}At the reference frame of viewpoint direction, wherein, DCMB _{L, t}Expression current block MB _{R, t}At F _{L, t}In the parallax compensation piece, MCMB _{L, t-1}Expression DCMB _{L, t}At F _{L, t-1}In motion compensation block, MCMB _{R, t-1}Expression current block MB _{R, t}At F _{R, t-1}In motion compensation block, DCMB _{L, t-1}Expression MCMB _{R, t-1}At F _{L, t-1}In the parallax compensation piece.

The motion vector of " stereo-picture to " and difference vector relation can be expressed from the next,

δ＝‖mv _r，t(MB _r，t)+dv _r，t-1(MCMB _r，t-1)-dv _r，t(MB _r，t)-mv _l，t(DCMB _l，t)‖

Wherein, ‖ v ‖ represents the norm of v, and δ is the model error of solid-kinematic constraint model, mv _{R, t}(MB _{R, t}) be current block MB _{R, t}The optimal motion vector, dv _{R, t}(MB _{R, t}) be current block MB _{R, t}Optimum difference vector, dv _{R, t-1}(MCMB _{R, t-1}) be MCMB _{R, t-1}Difference vector, mv _{L, t}(DCMB _{L, t}) be DCMB _{L, t}Motion vector.

And if only if MCMB _{L, t-1}And DCMB _{L, t-1}For same object when the true convergent point on different 3D surface, the δ in the following formula equals 0, that is,

mv _r，t(MB _r，t)+dv _r，t-1(MCMB _r，t-1)＝dv _r，t(MB _r，t)+mv _l，t(DCMB _l，t)

The present invention is directed to the encoding block design motion vector of all frames of the right viewpoint of three-dimensional video-frequency except that first frame and the method for quick predicting of difference vector, right viewpoint first frame still adopts full-search algorithm to guarantee search precision.Fig. 3 is the flow chart of the inventive method, is divided into initialization, adjusts and revise search box size, iterative search and four steps of abort criterion.Suppose the macro block MB in the right viewpoint t time chart picture _{R, t}Be current block, i represents the iteration step number, and δ is the model error of solid-kinematic constraint model, mv _{R, t}(MB _{R, t}) and dv _{R, t}(MB _{R, t}) represent the optimal motion vector of current block and the optimum difference vector of current block respectively, method step of the present invention is as follows:

The first step, initialization:

1) makes i=0, δ=0;

2) by candidate set

With

Determine the motion-vector search starting point of current block

Difference vector search starting point with current block

Wherein, mv _a/ dv _a, mv _b/ dv _bAnd mv _c/ dv _cRepresent left side piece a, upper block b that current block is adjacent and motion or the difference vector of upper right c respectively, mv _MedAnd dv _MedThe median vector of representing the motion vector and the difference vector of current block respectively, median vector mv _MedHorizontal component and vertical component equal mv respectively _a, mv _bAnd mv _cThe intermediate value of horizontal component and vertical component, median vector dv _MedHorizontal component and vertical component equal dv respectively _a, dv _bAnd dv _cThe intermediate value of horizontal component and vertical component, mv _{L, t}Be the motion vector of current block piece identical in the viewpoint direction reference frame, dv with the current block position _{R, t-1}Be the difference vector of current block piece identical in the time orientation reference frame, as shown in Figure 4 with the current block position.Because the adjacent block of current block has been finished coding, each adjacent block all has only motion vector or difference vector, and for example, only there is mv in the adjacent block a for the current block left side _aPerhaps dv _a, therefore, if a encodes mv in the disparity estimation mode _aJust do not exist.In this case, we replace mv with (0,0) _aCalculate median vector mv _MedRelatively the rate distortion costs of each candidate's predictive vector is chosen motion and the difference vector search starting point as the current block that make the rate distortion costs minimum respectively

With

3) respectively with the motion-vector search starting point of current block

Difference vector search starting point with current block

Be the center, delimit a RSR _MIN* RSR _MINThe correction search window, in this search window, do the vector correction, obtain the motion vector of revised current block and difference vector search starting point With

Preserve the rate distortion costs of the motion-vector search starting point of revised current block, note is done

Preserve the rate distortion costs of the difference vector search starting point of revised current block, note is done Make i=i+1.

Second step is according to the size of following formula adjustment correction search window RSR.

$\mathrm{RSR}=\left\{\begin{array}{cc}{\mathrm{RSR}}_{\mathrm{MIN}}& \mathrm{\&delta;}<{\mathrm{<figure-callout\; id="1"\; label="T"\; filenames="HDA0000043764300000021.png,HDA0000043764300000041.png"\; state="\{\{state\}\}">T</figure-callout>}}_1\\ {\mathrm{RSR}}_{\mathrm{MIN}}+\frac{\mathrm{\&delta;}-T_1}{T_2-T_1}({\mathrm{RSR}}_{\mathrm{MAX}}-{\mathrm{RSR}}_{\mathrm{MIN}})& {\mathrm{<figure-callout\; id="1"\; label="T"\; filenames="HDA0000043764300000021.png,HDA0000043764300000041.png"\; state="\{\{state\}\}">T</figure-callout>}}_1\&le;\mathrm{\&delta;}\&le;{\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA0000043764300000021.png,HDA0000043764300000041.png"\; state="\{\{state\}\}">T</figure-callout>}}_2\\ {\mathrm{RSR}}_{\mathrm{MAX}}& \mathrm{\&delta;}>{\mathrm{<figure-callout\; id="2"\; label="T"\; filenames="HDA0000043764300000021.png,HDA0000043764300000041.png"\; state="\{\{state\}\}">T</figure-callout>}}_2\end{array}\right.$

Wherein, RSR _MINBe the correction search window (being generally 2 * 2 pixels) of minimum, RSR _MAXBe the correction search window of maximum (get the search box size of full-search algorithm usually, promptly 96 * 96), T ₁And T ₂Represent two threshold value (T ₁＜T ₂), be used for controlling the size of revising search window, when model error δ greater than threshold value T ₂The time (T ₂Be generally 20), current block is the motion occlusion area probably, needs to adopt big correction search window RSR _MAXGuarantee search precision; When model error δ less than threshold value T ₁The time, T ₁Usually get 5, the motion/disparity vector that the current block that obtains in the current iteration process is described is very near the optimal motion/difference vector of current block, so, adopt less correction search window RSR _MIN, just be enough to guarantee search precision; Otherwise, design the correction search window of a self adaptation size according to the size of model error δ.

The 3rd step, iterative search procedures:

1) by

Calculate the difference vector prediction initial value of current block in the i time iterative process

Wherein,

The motion vector of representing the current block that obtains in the i-1 time iterative process;

The difference vector of the encoding block that expression is covered by the motion compensation block of current block in the time orientation reference frame, u are represented the number of the encoding block that is capped, as shown in Figure 5, will make

Minimum

As the difference vector of the motion compensation block of current block in the time orientation reference frame in the i time iterative process, note is done

It is the motion vector of the parallax compensation piece of current block in the viewpoint direction reference frame that obtain in the i-1 time iterative process.

With Be the center, delimit the search window of a RSR * RSR, in this search window, carry out the vector correction, obtain the difference vector of revised current block in the i time iterative process

Preserve the rate distortion costs of the difference vector of revised current block in the i time iterative process, note is done

2) by

Calculate the motion-vector prediction initial value of current block in the i time iterative process

Wherein,

The difference vector of revised current block in the i time iterative process representing to obtain in the 3rd step step 1),

The difference vector of the encoding block that expression is covered by the parallax compensation piece of current block in the viewpoint direction reference frame, v are represented the number of the encoding block that is capped, as shown in Figure 6, will make

Minimum

As the motion vector of the parallax compensation piece of current block in the i time iterative process in the viewpoint direction reference frame, note is done

The difference vector of the motion compensation block of current block in the time orientation reference frame in the i time iterative process representing to obtain in the 3rd step step 1).

With

Be the center, delimit the search window of a RSR * RSR, in this search window, carry out the vector correction, obtain the motion vector of revised current block in the i time iterative process Preserve the rate distortion costs of the motion vector of revised current block in the i time iterative process, note is done

The 4th step, abort criterion:

The rate distortion costs of representing the motion vector of the i time revised current block of iterative process,

The rate distortion costs of representing the motion vector of the i-1 time revised current block of iterative process,

The rate distortion costs of representing the difference vector of the i time revised current block of iterative process,

The rate distortion costs of representing the difference vector of the i-1 time revised current block of iterative process, if

And

Then make the motion vector and the difference vector of revised current block in the i-1 time iterative process respectively With

Optimal motion vector mv as current block _{R, t}(MB _{R, t}) and the optimum difference vector dv of current block _{R, t}(MB _{R, t}), the finishing iteration search procedure.Otherwise, make i=i+1, by Recomputate and upgrade δ, jump to step B, wherein,

With The difference vector of representing revised current block in the motion vector of revised current block in the i time iterative process and the i time iterative process respectively,

The difference vector of representing the motion compensation block of current block in the time orientation reference frame in the i time iterative process,

The motion vector of the parallax compensation piece of current block in the viewpoint direction reference frame in the i time iterative process.

In order to check the performance of method proposed by the invention, method of the present invention and all direction search method are compared.Experiment porch is JMVM8.0, estimation and disparity estimation search box size are 32 * 32, macro block mode is Inter16 * 16, from " Ballroom ", " Exit ", " Vassar ", " Racel ", the viewpoint 0 of " Rena " sequence, respectively get 100 two field pictures in the viewpoint 1 respectively as left viewpoint and right viewpoint, left side viewpoint adopts traditional h.264 coding method to encode earlier, and right viewpoint adopts stereo scopic video coding iteratively faster searching method of the present invention to encode, and each sequence resolution is 640 * 480.All experiments are being configured to Intel (R) Core (TM) 2 Extreme X9650 2.99GHzCPU, and the PC of 4GB RAM goes up independent the execution.

Table 1 is that the coding result of algorithms of different compares.Fig. 7 is the rate distortion curve signal of " Ballroom " sequence algorithms of different.As can be seen, this paper algorithm is compared with the full search of JMVM, the variation of Y-PSNR at-0.33dB between+the 0.01dB, the excursion of average bit rate is between-10.93% to+0.07%, basically do not see the decline of coding quality, and this paper algorithm can be saved the scramble time more than 96%.

The coding result of table 1 algorithms of different and JMVM full-search algorithm are relatively

Claims (4)

1. a stereo scopic video coding iteratively faster searching method makes the macro block MB in the right viewpoint t time chart picture _{R, t}Be current block, i is the iteration step number, and δ is the model error of solid-kinematic constraint model, mv _{R, t}(MB _{R, t}) expression current block the optimal motion vector, dv _{R, t}(MB _{R, t}) expression current block optimum difference vector,

Be the rate distortion costs of the motion vector of the i time revised current block of iteration,

Be the rate distortion costs of the motion vector of the i-1 time revised current block of iteration,

Be the rate distortion costs of the difference vector of the i time revised current block of iteration,

Be the rate distortion costs of the difference vector of the i-1 time revised current block of iteration, it is characterized in that may further comprise the steps:

1.1, initialization; Determine the motion-vector search starting point of current block

Difference vector search starting point with current block

Obtain the motion-vector search starting point of revised current block

Difference vector search starting point with revised current block

Preserve the motion of revised current block and the rate distortion costs of difference vector search starting point;

1.2, according to the size of following formula adjustment correction search window RSR,

$\mathrm{RSR}=\left\{\begin{array}{cc}{\mathrm{RSR}}_{\mathrm{MIN}}& \mathrm{\&delta;}<T_1\\ {\mathrm{RSR}}_{\mathrm{MIN}}+\frac{\mathrm{\&delta;}-T_1}{T_2-T_1}({\mathrm{RSR}}_{\mathrm{MAX}}-{\mathrm{RSR}}_{\mathrm{MIN}})& T_1\&le;\mathrm{\&delta;}\&le;T_2\\ {\mathrm{RSR}}_{\mathrm{MAX}}& \mathrm{\&delta;}>T_2\end{array}\right.$

Wherein, T ₁And T ₂Represent two threshold value (T ₁＜T ₂), RAR _MINBe minimum correction search window, RSR _MAXBe the maximum modified search window;

1.3, iterative search procedures; Determine the difference vector prediction initial value of the i time iteration current block

Carry out the vector correction, obtain the difference vector of the i time revised current block of iteration

Determine the motion-vector prediction initial value of the i time iteration current block

Carry out the vector correction, obtain the motion vector of the i time revised current block of iteration Preserve the rate distortion costs of the motion vector and the difference vector of the i time revised current block of iteration;

1.4, abort criterion: if

And

Then make the motion vector of the i-1 time revised current block of iteration

And difference vector Respectively as the optimal motion vector mv of current block _{R, t}(MB _{R, t}) and the optimum difference vector dv of current block _{R, t}(MB _{R, t}), the finishing iteration search procedure, otherwise, make i=i+1, by Recomputate and upgrade δ, jump to step 1.2, wherein,

With

Represent the motion vector of the i time revised current block of iteration and the difference vector of the i time revised current block of iteration respectively;

Wherein,

The difference vector of representing the motion compensation block of the i time iteration current block in the time orientation reference frame,

The motion vector of the parallax compensation piece of the i time iteration current block in the viewpoint direction reference frame.

2. stereo scopic video coding iteratively faster searching method according to claim 1 is characterized in that described step 1.1 comprises:

2.1, make i=0, δ=0;

2.2, the motion-vector search starting point of described definite current block

Difference vector search starting point with current block

Pass through candidate set

With Obtain;

Wherein, mv _a/ dv _a, mv _b/ dy _bAnd mv _c/ dv _cRepresent left side piece a, upper block b that current block is adjacent and motion or the difference vector of upper right c respectively, mv _MedAnd dv _MedRepresent the median vector of current block motion vector and the median vector of current block difference vector respectively, mv _{L, t}Be the motion vector of current block piece identical in the viewpoint direction reference frame, dv with the current block position _{R, t-1}Difference vector for current block piece identical in the time orientation reference frame with the current block position;

2.3, the described motion-vector search starting point that obtains revised current block

Difference vector search starting point with revised current block

Respectively with the motion-vector search starting point of current block

Difference vector search starting point with current block

Be the center, delimit a RSR _MIN* RSR _MINThe correction search window, in this search window, do the vector correction and obtain; The rate distortion costs of the motion-vector search starting point of the revised current block of described preservation, note is done

The rate distortion costs of the difference vector search starting point of the revised current block of described preservation, note is done Make i=i+1.

3. stereo scopic video coding iteratively faster searching method according to claim 1 is characterized in that the threshold value T in the described step 1.2 ₁Be 5, threshold value T ₂Be 20, RSR _MINBe 2, RSR _MAXBe 96.

4. stereo scopic video coding iteratively faster searching method according to claim 1 is characterized in that described step 1.3 comprises:

4.1, the difference vector of the i time iteration current block prediction initial value

By

Calculate;

Wherein,

The motion vector of representing the current block that the i-1 time iteration obtains; The difference vector of the encoding block that expression is covered by the motion compensation block of current block in the time orientation reference frame, u are represented the number of the encoding block that is capped, will make

Minimum

As the difference vector of the motion compensation block of the i time iteration current block in the time orientation reference frame, note is done

It is the motion vector of the parallax compensation piece of current block in the viewpoint direction reference frame that obtain of the i-1 time iteration;

With

Be the center, delimit the search window of a RSR * RSR, in this search window, carry out the vector correction, obtain the difference vector of the i time revised current block of iteration

Preserve the rate distortion costs of the difference vector of the i time revised current block of iteration, note is done

4.2, the motion-vector prediction initial value of the i time iteration current block

By

Calculate;

Wherein,

The difference vector of the i time revised current block of iteration that obtains in the expression step 4.1,

The difference vector of the encoding block that expression is covered by the parallax compensation piece of current block in the viewpoint direction reference frame, v are represented the number of the own encoding block that is capped, will make

Minimum

As the motion vector of the parallax compensation piece of the i time iteration current block in the viewpoint direction reference frame, note is done

The difference vector of the motion compensation block of the i time iteration current block in the time orientation reference frame that obtains in the expression step 4.1;

With Be the center, delimit the search window of a RSR * RSR, in this search window, carry out the vector correction, obtain the motion vector of the i time revised current block of iteration

Preserve the rate distortion costs of the motion vector of the i time revised current block of iteration, note is done

Images