CN107623858A - A kind of 3D video fast encoding methods that rule is skipped based on adaptive segmentation - Google Patents

Abstract

The invention discloses a kind of 3D video fast encoding methods that rule is skipped based on adaptive segmentation, the problem of higher calculating SVDC values of complexity have selectively been skipped in coding depth graphic sequence, belong to HEVC field of video encoding.It is characterized in that, the region that will not cause to synthesize viewpoint distortion in depth map is determined according to texture smoothness criterion and the distortion criterion of depth zero first, then carry out skipping rule based on adaptive segmentation to skip above-mentioned zero distortion zone in advance before SVDC is calculated in depth map, terminate the calculating process in the region, the method for being finally based on motion vector analyzes the interval that the forced coding pattern of each sequence to be encoded and optimal segmentation are skipped, and is effectively reduced encoder complexity.The present invention terminates the process that original coding method all carries out SVDC calculating to each coding unit in advance, on the premise of keeping coding quality constant, can averagely reduce by 21.711% scramble time.

Description

A kind of 3D video fast encoding methods that rule is skipped based on adaptive segmentation

Technical field

The present invention relates to the video coding technique based on HEVC, and in particular to a kind of 3D (three dimensional) regards The Fast video coding method of rule is skipped in frequency depth map encoding based on adaptive segmentation.

Background technology

In recent years, with the rapid development of multimedia technology, 3 D video is by academia, industrial quarters, particularly film The attention of industry, home entertaining industry.3D-HEVC standards support multi-view point video plus depth figure (Multi-view video plus Depth, MVD) video format coding, decoding end utilizes depth image rendering technique (Depth Image based Rendering, DBIR) the new virtual view of synthesis.Because depth map is used for the synthesis of virtual view, so coding depth figure Distortion can directly result in the distortion of virtual view.In order to consider the quality of virtual view while coding depth figure, based on rate The depth map encoding decision process of distortion optimizing method need to also consider virtually to regard while depth map itself coding distortion is calculated The distortion of point.Therefore, 3D-HEVC replaces the rate distortion during original depth-map coding mode decision excellent with synthesis view distortion Change (Rate Distortion Optimization, RDO), this technology is referred to as VSO.

At present, there are many domestic and foreign scholars to be all placed on research emphasis how to reduce in VSO complexity.Document [1] A kind of allowable depth distortion model with fragmentation feature is proposed, by minimizing to the View synthesis distortion under bit rates Carry out the new rate-distortion model of establishment model decision-making and motion/disparity estimation.In addition, in order to characterize View synthesis quality, document [2] Consider distortion caused by Video coding, distortion caused by depth quantization and intrinsic geometric distortion simultaneously.However, these algorithms It is the distortion that synthesis viewpoint is measured using mathematical modeling, their result is approximation, can accurately not represent that synthesis regards The distortion of point.In order to be precisely calculated synthesis viewpoint distortion, it is proposed that SVDC methods come measure synthesis view in distortion variations.

SVDC can be used for the distortion for accurately measuring VSO, and it is defined as the difference of the distortion of two synthesis viewpoints, calculating process As shown in Figure 2.Wherein V represents the void synthesized by uncoded original texture image frame T and uncoded original depth image frame D Intend visual point image frame, in addition, by encoded texture image frame T ＇ depth image frame D ＇ and D ＂ with two width code segments respectively Two width virtual visual point image frames of synthesis are denoted as V ＇ and V ＂, but D ＇ are also included on the basis of D ＂ and currently treated decision-making depth map encoding The depth information of distortion in block；By V ＇ and V ＂ respectively with V calculate squared difference and be designated as SSD ＇ and SSD ＂, finally calculate SSD ＇ and SSD ＂ difference, obtain SVDC value.

SVDC introducing will necessarily bring huge encoder complexity, in order to accelerate the calculating process of SVDC models, 3D-HEVC terminates some unnecessary SVDC calculating process using a kind of ES methods in advance.Current HTM-16.0 is to use Unnecessary encoding block is skipped so as to reduce the SVDC calculating time based on block and based on capable ES decision-making modes.ES methods Core skip condition is：If current depth value is identical with the difference vector that original depth value is calculated, the depth value is not It can cause to synthesize viewpoint distortion, can directly skip SVDC calculating.

But when calculating SSD ＂, each decision-making candidate can treat that decision-making encoding block encodes to obtain different predictions by current Block and reconstructed block so that encoder is required for calculating a SSD ＂ for each candidate, and this still has for coding side Very high computation complexity.Therefore, although ES methods can reduce encoder complexity to a certain extent, but still be improved Space.The present invention is based on HEVC standard, proposes a kind of Fast video coding method that rule is skipped based on adaptive segmentation, in advance The SVDC calculating process that original coding method is all carried out to each coding unit is terminated, coding quality can kept constant On the premise of, averagely reduce 21.711% scramble time.

It is attached：Bibliography

[1]Zhang Y,Kwong S,Hu SD,et al.Efficient multi-view depth coding Optimization based on allowable depth distortion in view synthesis.IEEE TRANSACTIONS ON IMAGE PROCESSING,NOV 2014, 23(11):4879-4892.

[2]Liu YW,Huang QM,Ma SW,et al.Joint video/depth rate allocation for 3D video coding based on view synthesis distortion model.SIGNAL PROCESSING- IMAGE COMMUNICATION,SEP 2009,24(8): 666-681.

[3]Ma SW,Wang SQ,Gao W.Low complexity adaptive view synthesis optimization in HEVC based 3D video coding.IEEETRANSACTIONS ON MULTIMEDIA,JAN 2014,16(1):266-271.

The content of the invention

It is an object of the present invention to it is directed to current three dimension high efficiency video encoding standard 3D-HEVC (3D extension of High efficiency video coding, 3D-HEVC) in depth map encoding use calculating synthesis viewpoint distortion variations The problem of (synthesized view distortion change, SVDC) complexity is high, it is proposed that one kind is based on adaptive View Synthesis optimization (view synthesis optimization, VSO) method of rule is skipped in segmentation.

In order to solve the above technical problems, the technical scheme that the present invention takes is：

A kind of 3D video fast encoding methods that rule is skipped based on adaptive segmentation, first with texture smoothness criterion The region that will not cause to synthesize viewpoint distortion in depth map is determined with the distortion criterion of depth zero, then carries out SVDC in depth map Rule is skipped to skip above-mentioned zero distortion zone in advance based on adaptive segmentation before calculating, terminates the calculating process in the region, most The method based on motion vector analyzes the interval that the forced coding pattern of each sequence to be encoded and optimal segmentation are skipped afterwards, effectively Ground reduces encoder complexity, comprises the following steps that：

1.1st, it is N x N to choose size, and coding mode is any one encoding block for treating decision-making coding mode；

1.2nd, it is detected successively using based on block and based on capable ES methods, if the encoding block meets to be based on block Or based on capable ES conditions, then the SVDC processes of whole encoding block terminate in advance, and transfer to judge whether next pixel column meets Based on block or based on capable ES conditions；

If the 1.3, a certain pixel column is unsatisfactory for, the judgement of texture smoothness criterion and the distortion criterion of depth zero is carried out, if Meet decision condition, then the SVDC of current region is set to 0, otherwise calculate the SVDC values of current region.

In the 3D video fast encoding methods provided by the present invention that rule is skipped based on adaptive segmentation, described line The decision process of slippery criterion is patted, is comprised the steps：

2.1st, the decision condition of texture smoothness criterion is, | l_i-l_i+1|≤T；

2.2nd, the threshold value T in the texture smoothness criterion in calculation procedure 2.1, calculation formula areWherein l_i,jAnd l_i+1,jIt is the pixel pair of a horizontal line, N_hAnd N_W-1Represent the big of each module Small, IntraXXBlockNum represents one kind in 35 kinds of intra-frame encoding modes.

In the 3D video fast encoding methods provided by the present invention that rule is skipped based on adaptive segmentation, described base The forced coding pattern of each sequence to be encoded is analyzed in the method for motion vector and optimal segmentation skips the process at interval, including Following step：

3.1st, the main moving direction of main movement object in current video sequence, the intraframe coding to be selected from 35 kinds are observed 5 optimal patterns are selected in pattern as candidate pattern；

3.2nd, 5 candidate patterns selected in step 3.1 are tested, passes through the BD rate in observation experiment result Final intra-frame encoding mode is determined with PSNR values；

3.3rd, the selection at optimal segmentation interval be by standard test sequences Balloons, Kendo, Newspaper, Poznan_Hall2, PoznanStreet } carry out test of many times and draw, the gap size generally chosen have 1,2,3,4, 5 }, by the relatively more required scramble time which judges at intervals of optimal spacing.

Brief description of the drawings

Fig. 1 is the flow chart of the inventive method；

Fig. 2 is SVDC (Synthesized View Distortion Change) calculating process block diagram；

Fig. 3 is 35 kinds of intra-frame encoding mode sketches；

Fig. 4 is the interval that forced coding pattern and optimal segmentation are skipped under the various criterion cycle tests selected；

Fig. 5 is the motion vector analysis figure of this sequence of Basketball；

Fig. 6 is the time of standard test sequences under the inventive method to save percentage column schematic diagram.

Embodiment

The present invention is further elaborated on below in conjunction with accompanying drawing.

Reference picture 1, a kind of 3D video fast encoding methods that rule is skipped based on adaptive segmentation, including：

1st, it is NxN to choose size, and coding mode is any one encoding block for treating decision-making coding mode；

2nd, it is detected successively using based on block and based on capable ES methods, if the encoding block meet based on block or Based on capable ES conditions, then the SVDC processes of whole encoding block terminate in advance, and transfer to judge whether next pixel column meets base In block or based on capable ES conditions；

If the 3, a certain pixel column is unsatisfactory for, the judgement of texture smoothness criterion and the distortion criterion of depth zero is carried out, if full Sufficient decision condition, then the SVDC of current region is set to 0, otherwise calculates the SVDC values of current region, calculating process is referring to Fig. 2；

4th, the method for being finally based on motion vector analyzes forced coding pattern and the optimal segmentation jump of each sequence to be encoded Super-interval, pattern to be encoded is referring to Fig. 3, and section gap to be selected is referring to Fig. 4.

Further, the decision process of described texture smoothness criterion, it is specially：

1) brightness value rebuild corresponding to current depth figure in texture maps between two adjacent pixels, i.e. horizontal pixel are utilized It is right, l_i,jAnd l_i+1,j, to define the smooth criterion of texture, such as formula | l_i-l_i+1Shown in |≤T；

2) threshold value T selection is to be used to determine currently to rebuild texture maps pixel to whether being in texture smooth area in step 1) Domain, in order to accurately find out this kind of zero distortion zone with the smooth criterion of texture, threshold value T needs targetedly to be set It is fixed；

Further, the threshold value T in the smooth criterion of texture proposed by the present invention can be after texture graph code, depth map volume It is calculated before code, such as formulaShown, its calculation procedure is as follows：

1) after a certain texture graph code, record all using a certain (DC patterns in generally use frame) in the texture maps Predictive mode is that the size of its final coding mode is N_w×N_hEncoding block；

2) recording step 1) in horizontal direction adjacent pixel in a certain encoding block, l_i,jAnd l_i+1,j, margin of measurement is calculated, Afterwards calculate full line absolute error and；

3) calculation procedure 2) in all pixels row absolute error and and average, obtain the average level of the encoding block Direction luminance difference；

4) average value is calculated to all horizontal direction luminance differences obtained by step 3) in the texture maps, and it is average with this Value defines the threshold value T in the smooth criterion of texture；

5) texture maps will update a subthreshold T after often encoding an I frame, therefore all in the cycle in same frame Image all uses same threshold value T；

Further, it is described in the 3D video fast encoding methods proposed by the present invention that rule is skipped based on adaptive segmentation Method based on motion vector analyze the forced coding pattern of each sequence to be encoded and optimal segmentation skips the process at interval, Comprise the steps：

1) referring to Fig. 5, the main moving direction of main movement object in current video sequence, the frame to be selected from 35 kinds are observed 5 optimal patterns are selected in interior coding pattern as candidate pattern；

2) 5 candidate patterns selected in step 1) are tested, by the BD rate in observation experiment result and PSNR values determine final intra-frame encoding mode；

3) by standard test sequences Balloons, Kendo, Newspaper, Poznan_Hall2, PoznanStreet } carry out test of many times and select optimal segmentation interval, the gap size generally chosen have 1,2,3,4, 5 }, referring to Fig. 6, by the relatively more required scramble time which judges at intervals of optimal spacing.

In order to examine the performance of method proposed by the invention, the method for the present invention is contrasted with original method. Experiment porch uses HTM16.0, cycle tests Balloons, Kendo, Newspaper, Poznan_Hall2 and PoznanStreet, specific test environment set as shown in table 1.

The test environment of table 1 is set

International testing standards of 3DV according to announcement to HTM original encoding method and set forth herein View Synthesis it is excellent Change method is compared, and scramble time results contrast is as shown in table 2.From table 1 it follows that set forth herein coding method have Effect reduces average complexity, average out to 21.711%.

The scramble time of the distinct methods of table 2 compares

Except being contrasted to the scramble time, also distortion performance is compared.As a result such as table, shown in 3.From table Can be seen that in 3, in addition to the BDBR of the synthesis viewpoint of Undo_Dancer this sequence is higher, set forth herein coding method base This maintains identical coding efficiency with original encoding method.

The coding method that table 3 proposes is compared with the distortion performance of original method

As can be seen from Table 2, in addition to the BDBR of the synthesis viewpoint of Balloons this sequence is higher, set forth herein volume Code method maintains identical coding efficiency with original encoding method substantially.

Claims (4)

1. a kind of 3D video fast encoding methods that rule is skipped based on adaptive segmentation, it is characterised in that comprise the steps of：

Step 1.1, selection size are N x N, and coding mode is any one encoding block for treating decision-making coding mode；

Step 1.2, using based on block and based on capable ES methods it is detected successively, if the encoding block meets to be based on block Or based on capable ES conditions, then the SVDC processes of whole encoding block terminate in advance, and transfer to judge whether next pixel column meets Based on block or based on capable ES conditions；

If step 1.3, a certain pixel column are unsatisfactory for, the judgement of texture smoothness criterion and the distortion criterion of depth zero is carried out, if Meet decision condition, then the SVDC of current region is set to 0, otherwise calculate the SVDC values of current region；

Step 1.4, the method for being finally based on motion vector analyze the forced coding pattern and optimal segmentation of each sequence to be encoded The interval skipped.

2. the 3D video fast encoding methods of rule are skipped based on adaptive segmentation as claimed in claim 1, it is characterised in that The decision process of described texture smoothness criterion, comprises the steps：

Step 2.1, the decision condition of texture smoothness criterion be, | l_i-l_i+1|≤T；

The threshold value T in texture smoothness criterion in step 2.2, calculation procedure 2.1, calculation formula areWherein l_i,jAnd l_i+1,jIt is the pixel pair of a horizontal line, N_hAnd N_W-1Represent the big of each module Small, IntraXXBlockNum represents one kind in 35 kinds of intra-frame encoding modes.

3. the 3D video fast encoding methods of rule are skipped based on adaptive segmentation as claimed in claim 2, it is characterised in that Step 2.2 is specially：

1) after a certain texture graph code, record and all in the texture maps use a certain predictive mode as its final coding mode Size be N_w×N_hEncoding block；

2) recording step 1) in horizontal direction adjacent pixel in a certain encoding block, l_i,jAnd l_i+1,j, margin of measurement is calculated, afterwards Calculate full line absolute error and；

3) calculation procedure 2) in all pixels row absolute error and and average, obtain the average level direction of the encoding block Luminance difference；

4) average value is calculated to all horizontal direction luminance differences obtained by step 3) in the texture maps, and determined with the average value Threshold value T in the adopted smooth criterion of texture；

5) texture maps will update a subthreshold T after often encoding an I frame, and all images in same frame in the cycle are all adopted With same threshold value T.

4. the 3D video fast encoding methods of rule are skipped based on adaptive segmentation as claimed in claim 3, it is characterised in that The described method based on motion vector analyzes the forced coding pattern of each sequence to be encoded and optimal segmentation skips interval Process, comprise the steps：

3.1st, the moving direction of moving object in current video sequence is observed, 5 are selected from intra-frame encoding mode to be selected most Good pattern is as candidate pattern；

3.2nd, 5 candidate patterns selected in step 3.1 are tested, by the BD rate in observation experiment result and PSNR values determine final intra-frame encoding mode；

3.3rd, the selection at optimal segmentation interval be by standard test sequences Balloons, Kendo, Newspaper, Poznan_Hall2, PoznanStreet } carry out test of many times and draw, the gap size generally chosen have 1,2,3,4, 5 }, by the relatively more required scramble time which judges at intervals of optimal spacing.