CN101340582A - Motion vector synthesizing method in resolution decreasing video code conversion - Google Patents

Abstract

The invention discloses a synthetic method for motion vectors in a degrading resolution video transcoding. Firstly, a first synthesized motion vector is calculated according to weight when the precisions of a plurality of input motion vectors are comparatively consistent or inconsistent; and the sum of absolute difference, bit number and lagrange multiplier which correspond to a median motion vector and a furthest motion vector apart from the first resultant motion vector in the adjacent motion vectors are calculated; and then a rate-distortion function value corresponding to each motion vector is calculated and the motion vector of the minimum rate-distortion function value is selected as a second resultant motion vector; finally, by adopting the second resultant motion vector as a center, a new search with a small window is carried out so as to obtain the final resultant motion vector. In the first synthesization of the invention, a weighted average is carried out to the motion vector based on the precision while a rate-distortion selection is carried out to the motion vector based on the precision in the second synthesization, thereby reducing errors in selecting the synthesization.

Description

Motion vector synthesizing method in a kind of resolution decreasing video code conversion

Technical field

The present invention relates to a kind of resolution decreasing video code conversion method, the motion vector synthesizing method in particularly a kind of resolution decreasing video code conversion belongs to multimedia video transcoding field.

Background technology

Video code conversion can be converted to another kind of form from a kind of form (as resolution, frame per second, code check etc.) with the video flowing that has compressed, with network and the terminal that adapts to different performance.As one of key technology of video code conversion, resolution decreasing video code conversion can be low resolution with compressed video by high resolution conversion by the space down-sampling, to adapt to the restriction of terminal display capabilities and disposal ability.In the resolution decreasing transcoding, a plurality of macro blocks in the high-definition picture need synthesize a macro block in the low-resolution image, and correspondingly a plurality of motion vectors also need synthetic new motion vector.Therefore motion vector becomes the key issue of resolution decreasing video code conversion.

Existing motion vector composition algorithm can be summarized as two kinds: first kind is to select, and second kind is summation.According to first kind of thinking, list of references 1 (Rajeev Kumar, Vasant Pati], An efficientmotion vector composition scheme for arbitrary frame down-sampling videotranscoder.IEEE Transactions on Circuits and Systems for VideoTechnology, 2006,16 (9): 1148～1152) proposed the leading vector in space and selected composition algorithm, from a plurality of input motion vectors, selected to have the motion vector of maximum coverage area as the output movement vector.According to second kind of thinking, list of references 2 (Jiao Wang, En-hui Yang, Xiang Yu, An efficientmotion estimation method for is video transcoding with spatialresolution conversion.In:Proceedings of 2007 IEEE InternationalConference on Multimedia and Expo H.264-based, Bei jing, China, 2007:444～447) proposed composition algorithm, a plurality of input motion vectors have been weighted summation according to linear model coefficients obtain the output movement vector based on linear degenerated mode.

Though above-mentioned algorithm makes moderate progress to the synthetic performance of motion vector, but exists a common problem: do not consider the accuracy of input motion vector, suppose that promptly the input motion vector is enough accurate.But in resolution decreasing video code conversion, this time is established and might not be set up.At first, when the sampling area of input motion vector correspondence less than the search area (for example when search for area be 16 * 16 and sampling area is 4 * 4) time, though it is the input motion vector is enough accurate with respect to the search area, then not necessarily enough accurate with respect to sampling area.Secondly, during the motion vector accuracy that video flowing the adopted tolerance before and after the transcoding different (for example when the H.264 video flowing of exporting behind the MPEG-2 video flowing of input before the transcoding and the transcoding adopt SAD (absolute difference with) and rate distortion function tolerance respectively) as the motion vector accuracy, though it is the input motion vector is enough accurate with respect to input video stream, then not necessarily enough accurate with respect to outputting video streams.

Summary of the invention

The purpose of this invention is to provide the motion vector synthesizing method in a kind of resolution decreasing video code conversion.

Technical scheme of the present invention may further comprise the steps:

(1) calculates input motion vector and weight between two kind situation between time lower, higher in accuracy.The precision of motion vector can be by the ratio S of its corresponding sampling area and search area _i/ S defines, ratio S _i/ S is higher greater than 3/4 accuracy, ratio S _i/ S is lower less than 1/4 accuracy, and other are then between low and between the two higher.Calculate input motion vector MV according to following formula _iAnd weight A between two kind situation between time lower, higher in accuracy _i, B _iAnd C _i:

$A_i=S_i/\frac{{\mathrm{SAD}}_i}{S_i}=S_i^2/{\mathrm{SAD}}_i$

$B_i=S_i\·\frac{{\mathrm{SAD}}_i}{S_i}={\mathrm{SAD}}_i$

C _i＝S _i

S wherein _i, SAD _iRepresent MV respectively _iCorresponding sampling area, absolute difference and, SAD _i/ S _iThe absolute difference of representation unit sampling area and, S _iCalculate SAD by the sampling ratio _iObtain by decoding transcoding input video stream.

(2) the resultant motion first time vector when more consistent according to a plurality of input motion vector of weight calculation accuracy.Calculate the first time resultant motion vector of a plurality of input motion vector accuracy when more consistent according to following formula:

${\mathrm{MV}}_A=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(A_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{A}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i^2/{\mathrm{SAD}}_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i^2/{\mathrm{SAD}}_i)}$

${\mathrm{MV}}_B=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(B_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{B}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}({\mathrm{SAD}}_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{\mathrm{SAD}}_i}$

${\mathrm{MV}}_C=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(C_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{C}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{S}_i}$

MV wherein _A, MV _B, MV _CRepresent all lower, the higher and resultant motion first time vector between two kinds of situations time of a plurality of input motion vector accuracy respectively, n represents the number of input motion vector;

(3) the resultant motion first time vector when inconsistent according to a plurality of input motion vector of weight calculation accuracy.Calculate the first time resultant motion vector MV of a plurality of input motion vector accuracy when inconsistent according to following formula _D:

${\mathrm{MV}}_D=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(D_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{D}_i}$

D wherein _iExpression input motion vector MV _iWeight when accuracy is inconsistent, formula specific as follows is calculated:

Wherein S represents MV _iCorresponding search area.

(4) calculate the vector of resultant motion for the first time correspondence absolute difference and, calculate MV respectively according to following formula _A, MV _B, MV _CAnd MV _DCorresponding absolute difference and SAD _A, SAD _B, SAD _CAnd SAD _D:

${\mathrm{SAD}}_A=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(A_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{A}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{S}_i^2}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i^2/{\mathrm{SAD}}_i)}$

${\mathrm{SAD}}_B=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(B_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{B}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{\mathrm{SAD}}_i^2}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{\mathrm{SAD}}_i}$

${\mathrm{SAD}}_C=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(C_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{C}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{S}_i}$

${\mathrm{SAD}}_D=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(D_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{D}_i}$

(5) calculate a current macro left side, last, upper right, upper left four adjacent motion vectors middle distances resultant motion for the first time vector farthest motion vector and the pairing absolute difference of intermediate value motion vector and;

(6) calculate the bit number of each motion vector correspondence according to motion vector, the motion vectors of current block and subsequent block;

(7) quantization step according to current block calculates Lagrange multiplier;

(8) according to the absolute difference of each motion vector correspondence and, bit number, Lagrange multiplier, calculate the rate distortion function value of each motion vector correspondence, calculate the rate distortion function value J (MV of each motion vector correspondence according to following formula _m):

J(MV _m)＝SAD _m+λ _MOTIONR _m

MV wherein _mCandidate's scope comprises the adjacent motion vectors MV in the transcoding output stream _N, MV _FWith the synthetic motion vector MV that the obtains first time _A, MV _B, MV _C, MV _DNumber.

(9) select to have the motion vector of minimum rate distortion function value as resultant motion second time vector;

(10) be the center with the resultant motion second time vector that obtains in the step (9), the motion vector that carries out wicket is heavily searched for and is obtained final resultant motion vector.The search window that motion vector is heavily searched for is [2,2].

The present invention and prior art contrast, effect is positive and tangible: the present invention is weighted on average motion vector based on accuracy in synthesizing for the first time, in synthesizing for the second time, motion vector is carried out rate distortion and select, thereby make the synthetic error of selection reduce based on accuracy.

Description of drawings

Accompanying drawing is the The simulation experiment result figure of Stefan cycle tests of the present invention.

Embodiment

Below be the concrete steps of the embodiment of the invention, what this embodiment carried out is that MPEG-2 arrives video code conversion H.264:

(1) precision of motion vector is by the ratio S of its corresponding sampling area and search area _i/ S defines, ratio S _i/ S is higher greater than 3/4 accuracy, ratio S _i/ S is lower less than 1/4 accuracy, and other are then between low and between the two higher.Calculate input motion vector MV respectively according to following formula _iAnd weight A between lower, higher two kind situations between time lower, higher in accuracy _i, B _iAnd C _i:

$A_i=S_i/\frac{{\mathrm{SAD}}_i}{S_i}=S_i^2/{\mathrm{SAD}}_i---\left(1\right)$

$B_i=S_i\·\frac{{\mathrm{SAD}}_i}{S_i}={\mathrm{SAD}}_i---\left(2\right)$

C _i＝S _i (3)

S wherein _i, SAD _iRepresent MV respectively _iCorresponding sampling area, absolute difference and, SAD _i/ S _iThe absolute difference of representation unit sampling area and, S _iSAD can be calculated by the sampling ratio _iThen can obtain by decoding transcoding input video stream;

(2) calculate a plurality of input motion vector accuracy resultant motion first time vector when consistent respectively according to following formula:

${\mathrm{MV}}_A=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(A_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{A}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i^2/{\mathrm{SAD}}_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i^2/{\mathrm{SAD}}_i)}---\left(4\right)$

${\mathrm{MV}}_B=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(B_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{B}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}({\mathrm{SAD}}_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{\mathrm{SAD}}_i}---\left(5\right)$

${\mathrm{MV}}_C=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(C_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{C}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{S}_i}---\left(6\right)$

MV wherein _A, MV _B, MV _CRepresent all lower, the higher and resultant motion first time vector between two kinds of situations time of a plurality of input motion vector accuracy respectively, n represents the number of input motion vector;

(3) calculate the first time resultant motion vector MV of a plurality of input motion vector accuracy when inconsistent according to following formula _D:

${\mathrm{MV}}_D=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(D_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{D}_i}---\left(7\right)$

D wherein _iExpression input motion vector MV _iWeight when accuracy is inconsistent, formula specific as follows is calculated:

Wherein S represents MV _iCorresponding search area;

(4) calculate MV respectively according to following formula _A, MV _B, MV _CAnd MV _DCorresponding absolute difference and SAD _A, SAD _B, SAD _CAnd SAD _D:

${\mathrm{SAD}}_A=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(A_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{A}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{S}_i^2}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i^2/{\mathrm{SAD}}_i)}---\left(9\right)$

${\mathrm{SAD}}_B=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(B_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{B}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{\mathrm{SAD}}_i^2}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{\mathrm{SAD}}_i}---\left(10\right)$

${\mathrm{SAD}}_C=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(C_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{C}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{S}_i}---\left(11\right)$

${\mathrm{SAD}}_D=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(D_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{D}_i}---\left(12\right)$

(5) a current macro left side, last, upper right, upper left four adjacent motion vectors middle distances resultant motion vector first time motion vector and intermediate value motion vector farthest are designated as MV respectively _FAnd MV _F, and calculate MV _F, MV _FCorresponding absolute difference and SAD _F, SAD _F

(6) calculate the bit number R of each motion vector correspondence respectively according to following formula _m:

R _m＝R(MV _m-MV _p)+R(MV′ _m-MV′ _p(MV′ _m)) (13)

MV wherein _mThe motion vector of expression current block, MV _pThe motion vectors of expression current block, MV ' _mThe motion vector of expression subsequent block, MV ' _p(MV ' _m) represent that subsequent block is MV at the current block motion vector _mThe time motion vectors.For the number of coded bits of current block motion vector difference, because MV _mWith MV _pAll known, therefore the number of coded bits of difference can directly obtain by tabling look-up between the two.For the number of coded bits of subsequent block motion vector difference, MV ' _p(MV ' _m) can be according to current block motion vector MV _mAnd other adjacent motion vectors obtain MV ' _mCan obtain the number of coded bits of motion vector difference then by tabling look-up with the motion vector approximate substitution of subsequent block in the transcoding inlet flow;

(7) according to list of references 3 (Zhang Y, Gao W, Lu Y, et al.Joint source-channelrate-distortion optimization for is video coding over error-pronenetworks.IEEE Transactions on Multimedia.2007 H.264,9 (3): macro block Lagrange multiplier formula 445-454), calculate Lagrange multiplier λ by the current macro quantization step _MOTION

(8) calculate the rate distortion function value J (MV of each motion vector correspondence according to following formula _m):

J(MV _m)＝SAD _m+λ _MOTIONR _m (14)

MV wherein _mCandidate's scope comprises the adjacent motion vectors MV in the transcoding output stream _E, MV _FWith the synthetic motion vector MV that the obtains first time _A, MV _B, MV _C, MV _D

(9) select to have the motion vector of minimum rate distortion function value as resultant motion second time vector;

(10) be the center with resultant motion second time vector, carry out search window and heavily search for for the motion vector of [2,2] and obtain final resultant motion vector.

Realize effect: in the embodiment of video code conversion H.264, adopt method of the present invention that the Stefan cycle tests is tested at MPEG-2.Experimental result shows that the method for complexity of the inventive method and list of references 2 is suitable substantially; And the signal to noise ratio of the inventive method as shown in drawings, compares with all direction search method and improved about 1.2dB.

Claims (7)

1. the motion vector synthesizing method in the resolution decreasing video code conversion is characterized in that may further comprise the steps:

(1) calculates input motion vector and weight between two kind situation between time lower, higher in accuracy;

(2) the resultant motion first time vector when more consistent according to a plurality of input motion vector of weight calculation accuracy;

(3) the resultant motion first time vector when inconsistent according to a plurality of input motion vector of weight calculation accuracy;

(4) calculate the vector of resultant motion for the first time correspondence absolute difference and;

(5) calculate a current macro left side, last, upper right, upper left four adjacent motion vectors middle distances resultant motion for the first time vector farthest motion vector and the pairing absolute difference of intermediate value motion vector and;

(6) calculate the bit number of each motion vector correspondence according to motion vector, the motion vectors of current block and subsequent block;

(7) quantization step according to current block calculates Lagrange multiplier;

(8) according to the absolute difference of each motion vector correspondence and, bit number, Lagrange multiplier, calculate the rate distortion function value of each motion vector correspondence;

(9) select to have the motion vector of minimum rate distortion function value as resultant motion second time vector;

(10) be the center with the resultant motion second time vector that obtains in the step (9), the motion vector that carries out wicket is heavily searched for and is obtained final resultant motion vector.

2. method according to claim 1 is characterized in that calculating input motion vector MV according to following formula in the step (1) _iAnd weight A between two kind situation between time lower, higher in accuracy _i, B _iAnd C _i:

$A_i=S_i/\frac{{\mathrm{SAD}}_i}{S_i}=S_i^2/{\mathrm{SAD}}_i$

$B_i=S_i\·\frac{{\mathrm{SAD}}_i}{S_i}={\mathrm{SAD}}_i$

C _i＝S _i

S wherein _i, SAD _iRepresent MV respectively _iCorresponding sampling area, absolute difference and, SAD _i/ S _iThe absolute difference of representation unit sampling area and, S _iCalculate SAD by the sampling ratio _iThen obtain by decoding transcoding input video stream.

3. method according to claim 1 is characterized in that in the step (2) calculating the first time resultant motion vector of a plurality of input motion vector accuracy when more consistent according to following formula:

${\mathrm{MV}}_A=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(A_i\·M{V}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{A}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i^2/{\mathrm{SAD}}_i\·M{V}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i^2/{\mathrm{SAD}}_i)}$

${\mathrm{MV}}_B=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(B_i\·M{V}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{B}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}({\mathrm{SAD}}_i\·M{V}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{\mathrm{SAD}}_i}$

${\mathrm{MV}}_C=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(C_i\·M{V}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{C}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i\·{\mathrm{MV}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{S}_i}$

MV wherein _A, MV _B, MV _CRepresent all lower, the higher and resultant motion first time vector between two kinds of situations time of a plurality of input motion vector accuracy respectively, n represents the number of input motion vector.

4. method according to claim 1 is characterized in that: calculate the first time resultant motion vector MV of a plurality of input motion vector accuracy when inconsistent according to following formula in the step (3) _D:

${\mathrm{MV}}_D=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(D_i\·M{V}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{D}_i}$

D wherein _iExpression input motion vector MV _iWeight when accuracy is inconsistent, formula specific as follows is calculated:

Wherein S represents MV _iCorresponding search area.

5. method according to claim 1 is characterized in that: step is calculated MV respectively according to following formula in (4) _A, MV _B, MV _CAnd MV _DCorresponding absolute difference and SAD _A, SAD _B, SAD _CAnd SAD _D:

${\mathrm{SAD}}_A=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(A_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{A}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{S}_i^2}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}\left(S_i^2{/\mathrm{SAD}}_i\right)}$

${\mathrm{SAD}}_B=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(B_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{B}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{\mathrm{SAD}}_i^2}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{\mathrm{SAD}}_i}$

${\mathrm{SAD}}_C=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(C_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{C}_i}=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(S_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{S}_i}$

${\mathrm{SAD}}_D=\frac{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}(D_i\·{\mathrm{SAD}}_i)}{\underset{i=0}{\overset{n}{\mathrm{\Σ}}}{D}_i}$

6. method according to claim 1 is characterized in that: the rate distortion function value J (MV that calculates each motion vector correspondence in the step (8) according to following formula _m):

J(MV _m)＝SAD _m+λ _MOTIONR _m

MV wherein _mCandidate's scope comprises the adjacent motion vectors MV in the transcoding output stream _E, MV _FWith the synthetic motion vector MV that the obtains first time _A, MV _B, MV _C, MV _DNumber.

7. method according to claim 1 is characterized in that: the search window that motion vector is heavily searched in the step (10) is [2,2].

Images