CN106254887A - A kind of deep video coding fast method - Google Patents

Abstract

The invention belongs to 3D field of video encoding, for realizing 3D HEVC deep video is carried out fast coding, ensureing to be effectively reduced in the case of coding efficiency does not has significantly sacrificing 3D HEVC computation complexity, saving the time of coding depth video.Take into full account deep video interframe gray value similarity, it is achieved effective deep video coding fast algorithm.The technical solution used in the present invention is, deep video coding fast method, and step is as follows: 1. gray value similar judgement 2.CU minimum code depth layer calculating 3. shifts to an earlier date decision-making based on the PU pattern that gray scale is similar.Present invention is mainly applied to 3D Video coding occasion.

Description

A kind of deep video coding fast method

Technical field

The invention belongs to 3D field of video encoding, relate to a kind of deep video coding utilizing gray value similarity and calculate quickly soon Method.

Background technology

Multiple views color video plus depth video (Multiview video plus depth, MVD) is a kind of effective 3D video data format.Depth map represents the geological information of scene, is the gray-scale map of only luminance component, each of which pixel Represent current object and the relative distance of camera plane in the 3 d space.Deep video is by the flat region being divided into by sharpened edge Territory forms, and has the characteristic different from color video.MVD video can be drawn out by utilizing rendering technique based on depth map Multiple virtual views.The 3D extended version of efficient video coding standard (High Efficiency Video Coding, HEVC) (3D-HEVC) new Predicting Technique and coding tools are introduced for coding MVD video, as disparity compensation prediction technology, motion are vowed Amount succession, depth model pattern (Depth Modeling Mode, DMM) and View Synthesis optimization (View Synthesis Optimization, VSO) etc..DMM pattern utilizes two kinds of different segmentation patterns that one depth block is divided into two non-rectangles Region, i.e. wedge shape segmentation.Each cut zone is represented by constant partition value.DMM can preferably represent the border of depth map And can obtain and predict more accurately, the most it is integrated in infra-frame prediction as candidate.Between component, coding tools kinematic parameter continues Holding (Motion Parameter Inheritance, MPI) utilizes the movable information of encoded color video to carry out depth map volume Code.

For deep video, 3D-HEVC is by introducing between viewpoint and between component, coding tools has reached well to encode effect Rate.But due to wedge shape segmentation decision-making, complicated rate-distortion optimization and the VSO process etc. of DMM, 3D-when coding depth video The computation complexity of HEVC is also in increase at double.Chinese scholars at present, for reducing 3D-HEVC computation complexity, saves Scramble time it is also proposed some effective methods.Tohidypour et al. proposes one and shifts to an earlier date adaptively in decision-making inter-frame The scheme of predictive mode search, the program mainly utilizes Movement consistency between viewpoint, predictive mode and rate distortion (Rate Distortion, RD) cost relatedness decreases 3D-HEVC computation complexity.Shen et al. proposes one rapidly and effectively Mode decision algorithm, by utilizing between Space-time domain adjacent encoder unit (Coding Unit, CU) and viewpoint and degree of depth interlayer pass Connection information saves the scramble time.Gu et al. proposes a kind of degree of depth intraframe coding quick DMM mode selection algorithm, and the method utilizes Likely model selection result terminates DMM full RD cost calculating process in advance thus obtains scramble time saving.

Summary of the invention

For overcoming the deficiencies in the prior art, it is contemplated that according to the gray value similarity of depth map, to the 3D-HEVC degree of depth Video carries out fast coding, is ensureing that being effectively reduced 3D-HEVC in the case of coding efficiency does not has significantly sacrificing calculates complexity Degree, saves the time of coding depth video.Take into full account deep video interframe gray value similarity, it is achieved effective deep video Coding fast algorithm.The technical solution used in the present invention is, deep video coding fast method, and step is as follows:

1. the similar judgement of gray value

The gray value similarity using depth map judges that can the CU depth layer information of encoded time domain reference frame be used for The reference information encoded as current CU, at current CU with the average gray value difference of corresponding CU in its reference frame less than similar thresholding Time, the coding depth of current CU is used for reference encoded with reference to CU depth layer scope, is limited the depth layer scope of self；CU maximum encodes Depth layer calculates

Based on gray value similarity calculated above, utilize the current CU's of correlation calculations of current CU and time domain reference CU Maximum coding depth layer, the maximum depth value of i.e. current CU quaternary tree recurrence, the maximum coding depth layer D of current CU_maxDefinition is such as Under:

$D_{max}=\left\{\begin{array}{cc}{D}_{ref},& if{S}_G=true\\ D_{org},& otherwise.\end{array}\right.$

Wherein D_orgIt is maximum coding depth layer in HTM primal algorithm, usually 3；D_refIt is corresponding in its time domain reference frame The maximum coding depth layer of CU, works as D_maxAfter determining, limit the division degree of depth of current CU thus avoid non-essential bigger depth layer RD cost calculate process；

2.CU minimum code depth layer calculates

First predetermined depth layer D of current CU is defined_preAs follows:

D_pre=α (D_left+D_above)+β·D_co

Wherein D_leftRepresent the depth capacity layer of left side CU adjacent for current CU, D_aboveRepresent top CU adjacent for current CU Depth capacity layer, D_coRepresenting the depth capacity layer of correspondence position CU in its time domain reference frame of current CU, α and β is weight factor；

The minimum-depth layer D of current CU_minAccording to D_preAnd S_GValue predict such as following formula:

3. shift to an earlier date decision-making based on the PU pattern that gray scale is similar

When current CU CU gray scale corresponding with its reference frame is similar, and select to merge Merge pattern or interframe with reference to CU When Inter2N × 2N pattern is as its optimum prediction mode, the mode decision process of current CU is by decision-making in advance；Excellent in rate distortion During changing (Rate Distortion Optimization, RDO), current CU needs pattern M of detection_cAccording to reference to CU's Pattern definition is as follows:

Wherein M_rRepresenting the optimum prediction mode of reference frame correspondence CU, All modes represents RDO mistake in HTM primal algorithm Journey needs the pattern of traversal.

The mark S of gray value similarity_GDefinition:

$S_G=\left\{\begin{array}{cc}true,& if|G_{cur}-G_{ref}|<TH\\ false,& otherwise.\end{array}\right.$

Wherein G_curIt is the average gray value of current CU, G_refIt is the average gray value of corresponding CU, TH in its time domain reference frame It it is the similar thresholding of gray scale.Grey similarity S is judged by similar thresholding TH_G；At G_curAnd G_refDifference less than similar thresholding During TH, then two CU have similar average gray value, S_GIt is that true i.e. two CU gray scales are similar；Otherwise S_GIt is false, i.e. two CU gray scales Dissimilar；

The feature of the present invention and providing the benefit that:

The present invention proposes a kind of new deep video based on gray value similarity coding quickly for deep video coding Algorithm.The present invention obtains considerable scramble time saving while ensureing coding efficiency, reduces 3D-HEVC and calculates complexity Degree.

Accompanying drawing illustrates:

Fig. 1 is the flow chart of technical scheme.

Detailed description of the invention

Deep video coding, for reducing 3D-HEVC computation complexity, is proposed a kind of new based on gray value phase by the present invention Like CU depth layer and the PU pattern high-speed decision method of property, thus save the depth map encoding time.Concrete technical scheme is divided into The following step:

1. the similar judgement of gray value

The gray value similarity using depth map judges that can the CU depth layer information of encoded time domain reference frame be used for The reference information encoded as current CU.At current CU with the average gray value difference of corresponding CU in its reference frame less than similar thresholding Time, the coding depth of current CU can be used for reference encoded with reference to CU depth layer scope, limit the depth layer scope of self.Gray value phase Mark S like property_GDefinition:

$S_G=\left\{\begin{array}{cc}true,& if|G_{cur}-G_{ref}|<TH\\ false,& otherwise.\end{array}\right.$

Wherein G_curIt is the average gray value of current CU, G_refIt is the average gray value of corresponding CU, TH in its time domain reference frame It it is the similar thresholding of gray scale.Grey similarity S is judged by similar thresholding TH_G.At G_curAnd G_refDifference less than similar thresholding During TH, then two CU have similar average gray value, S_GIt is that true i.e. two CU gray scales are similar；Otherwise S_GIt is false, i.e. two CU gray scales Dissimilar.

2.CU maximum coding depth layer calculates

Based on gray value similarity calculated above, utilize the current CU's of correlation calculations of current CU and time domain reference CU Maximum coding depth layer, the maximum depth value of i.e. current CU quaternary tree recurrence.The maximum coding depth layer D of current CU_maxDefinition is such as Under:

$D_{max}=\left\{\begin{array}{cc}{D}_{ref},& if{S}_G=true\\ D_{org},& otherwise.\end{array}\right.$

Wherein D_orgIt is maximum coding depth layer in HTM primal algorithm, usually 3；D_refIt is corresponding in its time domain reference frame The maximum coding depth layer of CU.Work as D_maxAfter determining, the division degree of depth of current CU can be limited thus avoid non-essential bigger deep The RD cost of degree layer calculates process.

3.CU minimum code depth layer calculates

First predetermined depth layer D of current CU is defined_preAs follows:

D_pre=α (D_left+D_above)+β·D_co

Wherein D_leftRepresent the depth capacity layer of left side CU adjacent for current CU, D_aboveRepresent top CU adjacent for current CU Depth capacity layer, D_coRepresenting the depth capacity layer of correspondence position CU in its time domain reference frame of current CU, α and β is weight factor.

The minimum-depth layer D of current CU_minAccording to D_preAnd S_GValue predict such as following formula:

4. shift to an earlier date decision-making based on the PU pattern that gray scale is similar

When current CU CU gray scale corresponding with its reference frame is similar, and select Merge or Inter 2N × 2N mould with reference to CU When formula is as its optimum prediction mode, the mode decision process of current CU is by decision-making in advance.During RDO, current CU needs inspection Pattern M surveyed_cAs follows according to the pattern definition with reference to CU:

Wherein M_rRepresenting the optimal mode of reference frame correspondence CU, All modes represents RDO process in HTM primal algorithm and needs Pattern to be traveled through.

Colored and the preferred forms of the deep video coding key present invention below by three viewpoints:

The present invention, based on the reference model HTM of 3D-HEVC, wherein regards three by the various patterns of 3D-HEVC definition Stippling normal complexion deep video encodes, and utilizes the former frame coding information of deep video to provide interframe for deep video present frame Reference information is used for predicting, described below:

1. the similar judgement of gray value

The gray value similarity using depth map judges that can the CU depth layer information of encoded time domain reference frame be used for The reference information encoded as current CU.First, the average gray value G of current CU CU corresponding with time domain reference frame is calculated_curWith G_refValue；Then pass through similar thresholding TH and judge grey similarity S_GGray value similarity, corresponding with its reference frame at current CU When the average gray value difference of CU is less than similar thresholding, the coding depth of current CU can be used for reference encoded with reference to CU depth layer scope, Limit the depth layer scope of self.The mark S of gray value similarity_GDefinition:

$S_G=\left\{\begin{array}{cc}true,& if|G_{cur}-G_{ref}|<TH\\ false,& otherwise.\end{array}\right.$

Wherein G_curIt is the average gray value of current CU, G_refIt is the average gray value of corresponding CU, TH in its time domain reference frame It it is the similar thresholding of gray scale.Specifically, at G_curAnd G_refDifference less than similar thresholding TH time, then explanation two CU have similar Average gray value, S_GIt is that true i.e. two CU gray scales are similar；Otherwise S_GBeing false, i.e. two CU gray scales are dissimilar.Similar thresholding TH is arranged It is 1, obtains by testing and weigh scramble time and coding efficiency in a large number.

2.CU maximum coding depth layer calculates

In the case of gray value is similar, depth layer information D of corresponding CU in time domain reference frame_refCan be as with reference to letter Breath instructs the coding of current CU.Therefore, based on the gray value similarity above calculated, the maximum coding depth layer D of current CU_max According to S_GValue calculate.Utilize the maximum coding depth layer of the current CU of correlation calculations of current CU and time domain reference CU, The maximum depth value of the most current CU quaternary tree recurrence.The maximum coding depth layer D of current CU_maxIt is defined as follows:

$D_{max}=\left\{\begin{array}{cc}{D}_{ref},& if{S}_G=true\\ D_{org},& otherwise.\end{array}\right.$

Wherein D_orgIt is maximum coding depth layer in HTM primal algorithm, usually 3；D_refIt is corresponding in its time domain reference frame The maximum coding depth layer of CU.Work as D_maxAfter determining, the depth capacity layer of i.e. current CU depth layer recursive traversal determines, passes through D_max The division degree of depth of current CU can be limited thus avoid the RD cost of non-essential bigger depth layer to calculate process.

3.CU minimum code depth layer calculates

Minimum-depth layer restrictive condition utilizes corresponding CU in spatial domain adjacent C U encoded for current CU and time domain reference frame Coding information calculates.First minimum-depth layer restrictive condition define predetermined depth layer D of current CU_preAs follows:

D_pre=α (D_left+D_above)+β·D_co

Wherein D_leftRepresent the depth capacity layer of left side CU adjacent for current CU, D_aboveRepresent top CU adjacent for current CU Depth capacity layer, D_coRepresenting the depth capacity layer of correspondence position CU in its time domain reference frame of current CU, α and β is weight factor, It is set to 0.25 and 0.5.

The minimum-depth layer D of current CU_minPredetermined depth layer D according to the current CU calculated_preWith current CU with at that time Gray value similarity S of territory reference frame correspondence CU_GValue predict such as following formula:

At S_GDuring for true value, under conditions of i.e. gray scale is similar, D_minAccording to D_preThe value that varies in size is different；In gray scale not phase Under the conditions of as, the minimum depth value of current CU will not be defined, D_min0 can be set to according to HTM primal algorithm.

4. shift to an earlier date decision-making based on the PU pattern that gray scale is similar

During RDO, current CU needs pattern M of detection_cOptimum prediction mode according to CU corresponding in time domain reference frame Determine.Under conditions of and reference CU similar to its time domain reference CU gray scale at current CU selects Merge pattern to be predicted, currently CU only detects the RD cost of Merge other pattern of mode skipping and calculates process；If selecting Inter 2N × 2N pattern to make with reference to CU During for optimum prediction mode, then current CU detects Merge pattern and Inter 2N × 2N pattern and skips the RD generation of other pattern Valency calculates.During RDO, current CU needs pattern M of detection_cAs follows according to the pattern definition with reference to CU:

Wherein M_rRepresenting the optimum prediction mode of corresponding CU in the time domain reference frame of current CU, it is former that All modes represents HTM In beginning algorithm, RDO process needs the pattern of traversal.

In sum, the invention provides a kind of deep video based on gray value similarity coding fast algorithm, three regard In the colored plus depth video of point, deep video component is limited by CU minimax depth layer scope based on gray value similarity And the decision making algorithm in advance of PU pattern, simplify CU depth layer traversal and the PU model prediction of original HTM platform deep video Process, thus decrease the deep video scramble time.

Claims (2)

1. a deep video coding fast method, is characterized in that, step is as follows:

1) the similar judgement of gray value

The gray value similarity using depth map judges that can the CU depth layer information of encoded time domain reference frame be used as The reference information of current CU coding, when current CU is less than similar thresholding with the average gray value difference of corresponding CU in its reference frame, The coding depth of current CU is used for reference encoded with reference to CU depth layer scope, limits the depth layer scope of self；

2) CU maximum coding depth layer calculates

Based on gray value similarity calculated above, utilize the maximum of the current CU of correlation calculations of current CU and time domain reference CU Coding depth layer, the maximum depth value of i.e. current CU quaternary tree recurrence, the maximum coding depth layer D of current CU_maxIt is defined as follows:

Wherein D_orgIt is maximum coding depth layer in HTM primal algorithm, usually 3；D_refIt is corresponding CU in its time domain reference frame Maximum coding depth layer, works as D_maxAfter determining, limit the division degree of depth of current CU thus avoid the RD of non-essential bigger depth layer Cost calculates process；

3) CU minimum code depth layer calculates

First predetermined depth layer D of current CU is defined_preAs follows:

D_pre=α (D_left+D_above)+β·D_co

Wherein D_leftRepresent the depth capacity layer of left side CU adjacent for current CU, D_aboveRepresent that adjacent for current CU top CU is Big depth layer, D_coRepresenting the depth capacity layer of correspondence position CU in its time domain reference frame of current CU, α and β is weight factor；

The minimum-depth layer D of current CU_minAccording to D_preAnd S_GValue predict such as following formula:

4) decision-making is shifted to an earlier date based on the PU pattern that gray scale is similar

When current CU CU gray scale corresponding with its reference frame is similar, and Merge or Inter 2N × 2N pattern is selected to make with reference to CU During for its optimum prediction mode, the mode decision process of current CU is by decision-making in advance；During RDO, current CU needs detection Pattern M_cAs follows according to the pattern definition with reference to CU:

Wherein M_rRepresenting the optimal mode of reference frame correspondence CU, All modes represents RDO process in HTM primal algorithm needs traversal Pattern.

2. a kind of deep video coding fast method as claimed in claim 1, is characterized in that, the mark S of gray value similarity_G It is defined as follows:

Wherein G_curIt is the average gray value of current CU, G_refBeing the average gray value of corresponding CU in its time domain reference frame, TH is ash Spend similar thresholding.Grey similarity S is judged by similar thresholding TH_G；At G_curAnd G_refDifference less than similar thresholding TH time, Then two CU have similar average gray value, S_GIt is that true i.e. two CU gray scales are similar；Otherwise S_GIt is false, i.e. two CU gray scale not phases Seemingly.