CN101001380A

CN101001380A - Motion Estimation Method for Video Coding

Info

Publication number: CN101001380A
Application number: CN 200610001211
Authority: CN
Inventors: 邱全成; 孙雷
Original assignee: Inventec Corp
Current assignee: Inventec Corp
Priority date: 2006-01-10
Filing date: 2006-01-10
Publication date: 2007-07-18

Abstract

The invention discloses a motion estimation method of video coding, which is used for calculating a shape motion vector aiming at a video object plane to be coded, and the method comprises the steps of firstly setting a first post-selection vector block group and a second post-selection vector block group corresponding to a block to be coded, detecting whether effective motion vectors exist in the first post-selection vector block group and the second post-selection vector block group, setting a shape motion vector predicted value of the block to be coded according to the number of the analyzed effective motion vectors, and then calculating whether the sum of absolute differences between the block corresponding to the shape motion vector predicted value and the block to be coded is smaller than a matching error standard threshold value, so as to obtain the shape motion vector of the block to be coded. The estimation method of the shape motion vector predicted value is more reasonable, the estimation result is relatively more accurate, the accuracy of the predicted value is greatly improved, the operation workload can be effectively reduced, the operation speed is improved, and the operation performance of the system is improved.

Description

The mobile estimating method of video coding

Technical field

The invention relates to a kind of mobile estimating method, particularly about a kind of mobile estimating method of being encoded at shape information in the video object plane.

Background technology

MPEG (Moving Picture Expert Group; Motion graphics panel of expert) be the working group that relevant mobile compressed encoding standard is formulated in specific responsibility, the standard of being formulated is a general international standard, mpeg standard just, and this standard partly is made up of video, audio frequency and system three.Wherein, MPEG-1 is videodisc (Video Compact Disc; VCD) video image compression standard; MPEG-2 coefficient word video video disc (Digital Video Disc; DVD)/the video image compression standard of super videodisc; MPEG-4 then is one of network video image compression standard, has characteristics such as compression ratio height, imaging be clear, compares with MPEG-1 and MPEG-2, and MPEG-4 pays attention to the interactivity and the flexibility of multimedia system more.

Moreover MPEG-4 is first compression method based on object (Object-Based).According to the MPEG-4 standard, each picture all is by video object plane (Video ObjectPlane; VOP) constitute, can be separately or associated form the object video in the scene (scene) (AV Object) encoded, stored and transmit operation.Object video is made up of shape (shape), texture (texture) and the information such as (motion) of moving in each video object plane, each video object plane in the MPEG-4 standard is to be divided into shape (Shape) information and texture (Texture) information is stored respectively, see also Figure 1A and Figure 1B, this Figure 1A and Figure 1B have shown corresponding shape information 11 pictures of a video object plane and texture information 12 pictures respectively, wherein, this video object plane corresponding shape information 11 is by a plurality of shape blocks (Binary Alpha Block; BAB) 110 formations, relatively, 12 of the texture informations of this video object plane correspondence are by 120 formations of a plurality of texture macro blocks (Macro Block), and comprise 4 texture blocks in each texture macro block 120.Moreover, in this shape information 11 in each shape block 110 and this texture information 12 position of each texture macro block 120 corresponding mutually.

Please cooperate and consult Fig. 2, if want the shape information 11 of encoded video object plane 1 to encode at this, then need each the shape block 110 in this shape information 11 is encoded one by one, particularly, be to want to define in the encoded video object plane 1 one at this to want encoded block 111, and at the previous moment video object plane that should want encoded video object plane 1 relatively, just seek in the reference video object plane 2 and be somebody's turn to do and want encoded block 111 immediate shape blocks as reference block 211, this search procedure promptly is called shape and moves estimation.

MPEG-4 coded data compression efficiency depends on and moves the accuracy of estimating, moves the accuracy of estimating and then depends on needed operation time.Please cooperate and consult Fig. 3 A, existing mobile estimating method is at first to set the back vector block of selecting of wanting encoded block 111, just adjacent area piece 113, adjacent area piece 115, adjacent area piece 117, adjacent area piece 123, adjacent area piece 125 and adjacent area piece 127, wherein, this adjacent area piece 113, adjacent area piece 115 and adjacent area piece 117 are that this wants to select vector-shaped block (16*16 pixel) after 111 correspondences of encoded block, and the motion-vector of this adjacent area piece 113, adjacent area piece 115 and adjacent area piece 117 is defined as (MVs1), (MVs2) and (MVs3) respectively; Adjacent area piece 123,127 of adjacent area piece 125 and adjacent area pieces are to select vectorial texture block (8*8 pixel) after this wants encoded block 111 pairing, and with this adjacent area piece 123, the motion-vector of adjacent area piece 125 and adjacent area piece 127 is defined as (MV1) respectively, (MV2) and (MV3), when the estimation operation is moved in execution, at first select vector lists { MVs1 after this, MVs2, MVs3, MV1, MV2, choose first effective motion-vector among the MV3}, as the shape motion vector prediction value of wanting encoded block 111 (Motion Vector Predictor for shape, hereinafter to be referred as MVPs), i.e. motion-vector shown in Fig. 3 B figure (MVs4 and MV4), thereafter, calculate this MVPs corresponding reference block 211 and be somebody's turn to do the absolute difference and (the Sumof Absolute Difference that want 111 of encoded block, hereinafter to be referred as SAD), and judge that whether SAD is less than match error threshold (industry specified standard value), if this SAD is less than match error threshold, MVPs was wanted the final shape motion-vector of encoded block 111 as this; When SAD greater than match error threshold, in a preset range, just be the center then with MVPs, ± 16 pixels are to search in the scope of radius, obtain final MVs.

The shortcoming of above-mentioned technology is to be, determining of MVPs is from candidate list { MVs1, MVs2, MVs3, MV1, MV2, choose first effective motion-vector among the MV3}, this method of estimation has bigger randomness, can't guarantee whether first selected effective motion-vector is appropriate, so can increase system's operand greatly, also can improve the complexity of computer hardware; In addition, the MVPs error that adopts above-mentioned method of estimation to calculate is bigger, and accuracy is lower; Moreover, if the SAD that calculates according to this MPVs correspondence is during greater than match error threshold, because of shape motion vector prediction value MPVs locating accuracy not high, then must ± 16 pixels in a big way in search for once more, this measure has not only increased the complexity of calculating, and it is more to expend time in, and then influences the fluency of real-time coding operation.

In sum, how to develop a kind of shape mobile estimating method, can realize maximum accuracy as a result with the computational complexity of minimum, the real important topic of needing research for the field relevant badly with video coding.

Summary of the invention

For overcoming the shortcoming of above-mentioned prior art, main purpose of the present invention is to provide a kind of mobile estimating method that improves the video coding of arithmetic speed.

Another object of the present invention is to be to provide a kind of mobile estimating method that improves the video coding of shape motion vector prediction value accuracy.

For reaching above-mentioned purpose and other relevant purpose, the invention provides a kind of mobile estimating method of video coding, the mobile estimating method of video coding of the present invention is at the video object plane that will encode (Video Object Plane; VOP), calculate its shape motion-vector (Motion Vectorof Shape; MVs), this method may further comprise the steps: (1) set want the encoded block correspondence first after select the vector block group after selecting vector block group and second; (2) detect to should want encoded block first after select whether there is effective motion-vector in the vector block group, if not, then enter step (3), if then enter step (4); (3) detect to should want encoded block second after select whether there is effective motion-vector in the vector block group, if not, promptly setting this, to want the shape motion vector prediction value of encoded block be 0, and enter step (5), if then enter step (4); (4) the effective motion-vector that arrives of analyzing and testing, and, the shape motion vector prediction value that this wants encoded block is set according to the effective motion-vector number that analyzes; (5) calculate this shape motion vector prediction be worth pairing block and should want between encoded block absolute difference with, and judge this absolute difference and whether less than the matching error level threshold value, if promptly draw this shape motion-vector of wanting encoded block and equal this shape motion vector prediction value; If not, then enter step (6); And (6) in a preset range, search for and this encoded block between absolute difference and be minimum shape motion vector prediction value, draw the shape motion-vector that this wants encoded block.

Wherein, should want encoded block is a shape block (Binary AlphaBlock in this video object plane; BAB).Moreover, this selects the vector block group after first is to want the shape block of encoded block to be constituted by a plurality of facing mutually in this, this selects the vector block group after second is to want the texture block of encoded block to be constituted by a plurality of facing mutually in this, and the size of this shape block is to be the 16*16 pixel, and the size of this texture block is to be the 8*8 pixel.Moreover, be to be the center in step (6), and in the zone of the pixel (pixel) of predetermined quantity, search for this shape motion vector prediction value (MVPs) that calculates.

This selects the vector block group after first is to want shape block (the Binary Alpha Block of encoded block by three adjacent to this; BAB) constitute, this selects the vector block group after second is to want the texture block of encoded block to be constituted by three adjacent to this.And it is further comprising the steps of in step (4): (4-1) number of the effective motion-vector of analysis, if effectively the number of motion-vector is 1, then enter step (4-2), if effectively the number of motion-vector is 2, then enter step (4-3), if effectively the number of motion-vector is 3, then enter step (4-4); (4-2) this being set wants the shape motion vector prediction value of encoded block to equal this effective motion-vector; (4-3) average that this shape motion vector prediction value of wanting encoded block equals these two effective motion-vectors is set; And (4-4) this shape motion vector prediction value of wanting encoded block is set and equals in the median of these three effective motion-vectors or the average one.

Compared with prior art, the shape motion vector prediction value that the mobile estimating method of video coding of the present invention not only estimates is more accurate, and when this shape motion vector prediction value can't satisfy standard-required, can the final shape motion-vector of less ground range searching, thus the speed of system's computing improved.

Description of drawings

Figure 1A and Figure 1B show shape information and the texture information that existing video coding processing back is generated for a video object plane (VOP) respectively;

Fig. 2 is the mobile estimation principle schematic diagram of explanation video coding;

Fig. 3 A be encoded block of the present invention and corresponding after select the schematic diagram of vector block 1;

Fig. 3 B be encoded block of the present invention and corresponding after select another schematic diagram of vector block; And

Fig. 4 is the operation workflow figure of the mobile estimating method of video coding of the present invention.

Embodiment

Fig. 4 is the operation workflow figure of the mobile estimating method of video coding of the present invention.Also please cooperate Figure 1A, Figure 1B, Fig. 2, Fig. 3 A and Fig. 3 B to describe the operation workflow of the mobile estimating method of video coding of the present invention in detail at this.In the present embodiment, this mobile estimating method is at the video object plane that will encode (Video Object Plane; VOP), calculate its shape motion-vector (Motion Vector of Shape; Be designated hereinafter simply as MVs), wherein, this shape information 11 is by the shape block of a plurality of 16*16 of being of a size of pixels (Pixel) (Binary Alpha Block; BAB) 110 form, and 12 of this texture informations are made of a plurality of texture macro blocks (Macro Block) 120 that are of a size of the 16*16 pixel, and also comprise 4 texture blocks that are of a size of the 8*8 pixel in each texture macro block 120.

As shown in the figure, execution in step S200 at first, set the back vector block of selecting of wanting the encoded block correspondence, wherein, should want encoded block 111 is to want a shape block in the encoded video object plane 1, and be to select the vector block group after selecting vector block group and second after wanting the back choosing vector of encoded block 111 correspondences to be divided into first this, and this selects the vector block group after first is to want the shape block of encoded block 111 to be constituted by a plurality of facing mutually in this, in the present embodiment, this selects the vector block group to comprise after first and lays respectively at the left side that this wants encoded block 111, the adjacent area piece 113 at top and top, right side, adjacent area piece 115 and adjacent area piece 117, it is respectively with MVs1, MVs2 and MVs3 indicate; This selects the vector block group after second is to want the texture block of encoded block 111 to be constituted by a plurality of facing mutually in this, see also Fig. 3 A, wherein, should want encoded block 111 correspondences first after select each the shape block that comprises in the vector block group, this adjacent area piece 113 just, adjacent area piece 115 and adjacent area piece 117 pairing texture macro blocks are to be divided into this texture macro block 122, texture macro block 124 and texture macro block 126, in the present embodiment, this selects after second the vector block group to be made of texture block corresponding in above-mentioned each texture macro block, just this selects the vector block group to comprise after second and lays respectively at the left side that this wants encoded block 111, the adjacent area piece 123 at top and top, right side, adjacent area piece 125 and adjacent area piece 127 constitute, can be respectively with MV1, MV2 and MV3 indicate (promptly as shown in Figure 3A).It should be noted that, of the present invention in the back number of selecting vector block of wanting encoded block 111 correspondences and present position not as limit, wherein, this selects the vector block group also can lay respectively at the left side that this wants encoded block 111 by four respectively after first, the top, the adjacent area piece 113 of top, right side and left side top, adjacent area piece 115, adjacent area piece 117 and adjacent area piece 119 are formed, this selects the vector block group then can lay respectively at the left side that this wants encoded block 111 by four respectively after second, the top, the adjacent area piece 123 of top, right side and left side top, adjacent area piece 125, adjacent area piece 127 and adjacent area piece 129 form (promptly shown in Fig. 3 B), then enter step S201.

In step S201, select vector block group { MVs1 after judging first, MVs2, whether there is effective motion-vector among the MVs3}, judge that just each wants whether to have effective motion-vector in the shape block of encoded block 111 adjacent to this, if not, then enters step S202, if then enter step S203.

In step S202, select the vector block group after first { there are not effective motion-vector in MVs1, MVs2 among the MVs3} judging this, select vector block group { MV1 after then then judging second, whether MV2 exists effective motion-vector among the MV3}, judge just each wants whether to have effective motion-vector in the texture block of encoded block 111 adjacent to this, if not, then enter step S207, if then enter step S203.

In step S203, analyze the number of effective motion-vector, specifically be to select vector block group { MVs1 after analyzing first, MVs2, select vector block group { MV1, MV2, the number of the effective motion-vector that exists among the MV3} behind the MVs3} or second, if effectively the number of motion-vector is 1, then enter step S204; If effectively the number of motion-vector is 2, then enter step S205; If effectively the number of motion-vector is 3, then enter step S206.

In step S204, MVPs is set equals this effective motion-vector, for example if analyze draw this select after first the vector block group MVs1, MVs2, among the MVs3} only MVs1 be effective motion-vector, then make MVPs=MVs1, then enter step S208.

In step S205, the average that MVPs equals these two effective motion-vectors is set, for example draw this and select vector block group { MVs1 after first if analyze, MVs2, it among the MVs3} effective motion-vector, then make (MVPs)=((MVs1)+(MVs2))/2, then enter step S208.

In step S206, MVPs is set equals in the median of these three effective motion-vectors or the average one, for example draw this and select vector block group { MVs1 after first if analyze, MVs2, MVs1 among the MVs3}, MVs2 and MVs3 are effective motion-vector, select vector block group { MVs1 after then taking out first, MVs2, median MVs2 between MVs3} is as MVPs, perhaps after first, select vector block group { MVs1, MVs2, (MVs1) of MVs3}, (MVs2) and (MVs3) calculate, calculating is obtained average (((MVs1)+(MVs2)+(MVs3))/3 just) as (MVPs), then enters step S208.

In step S207, select the vector block group all not have effective motion-vector owing to judge after this selects vector block group and second after first, this shape motion vector prediction value of wanting encoded block 111 (Motion Vector Predictor for shape then is set; Hereinafter to be referred as MVPs) be 0, in the present embodiment, should (MVPs) represent with two bit value, just, (MVPs)=(0,0), and enter step S208.

In step S208, in reference video object plane 2, find out the pairing block of MVPs, just reference block 211, and calculate this reference block 211 and be somebody's turn to do the absolute difference and (the Sum of Absolute Difference that want 111 of encoded block; Hereinafter to be referred as SAD), need the expositor, because the operation method of SAD is general video coding technique, and also non-technical characterictic of the present invention, so do not repeat them here, then enter step S209.

In step S209, whether judge SAD less than match error threshold, in the present embodiment, this match error threshold is an industry specified standard value, i.e. (16*AlphaTH) if not, then carries out step S210, if then carry out step S211.

In step S210, in default regional extent, it just is the center with MVPs, and for example to add the block that searches out and should want the SAD minimum of 111 of encoded block in 5 and the regional extent that to subtract 5 pixels be radius, and utilize computing formula to calculate this and want the final MVs of encoded block 111, in the present embodiment, this computing formula is a shape motion-vector difference=(MVs)-(MVPs).So, can carry out the higher video coding of accuracy according to the shape motion-vector difference that obtains, and can finish the treatment step of the mobile estimating method of video coding of the present invention.

In step S211, this is set wants the shape motion-vector MVs of encoded block 111 to equal MVPs, just with this MVPs as should want the final shape motion-vector of encoded block 111, and carry out video coding, and can finish the treatment step of the mobile estimating method of video coding of the present invention according to the shape motion-vector obtained.

As from the foregoing, mobile estimating method of the present invention is to select the vector district after each in the prior art Piece is integrated into corresponding block group according to type, and according to the effective movement that exists in each block group The number of vector, the shape motion-vector that utilizes different account forms to draw to want encoded block is pre-Measured value, compared with prior art in, only to select in the vector block first effectively mobile after each Vector is as the shape motion vector prediction value of wanting encoded block, shape motion-vector of the present invention The estimation mode of predicted value is more reasonable, and the estimation result is also relatively more accurate. In addition, by Accuracy in this shape motion vector prediction value improves greatly, thereby when this shape motion-vector When predicted value did not meet standard-required, range searching that then can be littler should want encoded block final The shape motion-vector, so, can effectively reduce the computing workload, improve arithmetic speed, Improved the operation performance of system.

Claims

1. a motion estimation method of video coding, for the video object plane to be coded, calculate its shape motion vector, it is characterized in that, the method comprises the following steps:

(1) Setting the first candidate vector block group and the second candidate vector block group corresponding to the coding block;

(2) Detect whether there is an effective motion vector in the first candidate vector block group corresponding to the coding block, if not, then proceed to step (3), if so, then proceed to step (4);

(3) Detect whether there is an effective motion vector in the second candidate vector block group corresponding to the block to be encoded, if not, set the predicted value of the shape motion vector of the block to be encoded to 0, and proceed to the step (5), if so, then proceed to step (4);

(4) Analyzing the detected effective motion vectors, and setting the shape motion vector prediction value of the block to be coded according to the number of effective motion vectors analyzed;

(5) Calculate the sum of absolute differences between the block corresponding to the predicted value of the shape motion vector and the block to be coded, and judge whether the sum of absolute differences is less than the matching error standard threshold, and if so, obtain the sum of the codes to be coded The shape motion vector of the block is equal to the predicted value of the shape motion vector; if not, proceed to step (6); and

(6) Search for the predicted value of the shape motion vector whose absolute difference sum with the encoding block is the smallest within a preset range, and obtain the shape motion vector of the block to be encoded.

2. The motion estimation method according to claim 1, wherein the block to be coded is a shape block in the video object plane.

3. The motion estimation method according to claim 1, wherein the first candidate vector block group is composed of a plurality of shape blocks adjacent to the block to be coded, and the second candidate The vector block group is composed of multiple texture blocks adjacent to the block to be coded.

4. The motion estimation method according to claim 2 or 3, wherein the size of the shape block is 16*16 pixels; the size of the texture block is 8*8 pixels.

5. The motion estimation method as claimed in claim 3, characterized in that, the step (4) further comprises the following steps:

(4-1) analyze the number of effective motion vectors, if the number of effective motion vectors is 1, then proceed to step (4-2), if the number of effective motion vectors is 2, then proceed to step (4-3), if the number of effective motion vectors is 3, then proceed to step (4-4);

(4-2) Setting the shape motion vector prediction value of the block to be coded to be equal to the effective motion vector;

(4-3) Setting the predicted value of the shape motion vector of the block to be coded equal to the mean value of the two valid motion vectors; and

(4-4) Set the predicted value of the shape motion vector of the block to be coded to be equal to one of the median value or the mean value of the three valid motion vectors.

6. The motion estimation method according to claim 1, wherein the first candidate vector block group is composed of three shape blocks adjacent to the block to be coded, and the second candidate The vector block group is composed of three texture blocks adjacent to the block to be coded.

7. The motion estimation method as described in the first patent application scope, it is characterized in that, in this step (6), it is based on the predicted value of the calculated shape motion vector as the center and with a predetermined number of pixels as the radius Search within the area.