CN105611299A - Motion estimation method based on HEVC - Google Patents
Motion estimation method based on HEVC Download PDFInfo
- Publication number
- CN105611299A CN105611299A CN201510994165.XA CN201510994165A CN105611299A CN 105611299 A CN105611299 A CN 105611299A CN 201510994165 A CN201510994165 A CN 201510994165A CN 105611299 A CN105611299 A CN 105611299A
- Authority
- CN
- China
- Prior art keywords
- current
- pixel
- carry out
- search
- hevc
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
- H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
- H04N19/51—Motion estimation or motion compensation
- H04N19/56—Motion estimation with initialisation of the vector search, e.g. estimating a good candidate to initiate a search
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/50—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding
- H04N19/503—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using predictive coding involving temporal prediction
- H04N19/51—Motion estimation or motion compensation
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a motion estimation method based on HEVC. The method comprises the steps of 1, performing integer pixel motion estimation on a current PU; 2, judging whether the fractional pixel motion estimation of the current PU is invalid, if yes, setting the integer pixel motion vector (MV) to be the optimal MV and skipping to the step 9, and otherwise, executing the step 3; 3, judging whether the current PU includes textural features, if yes, executing the step 4, and otherwise, adopting an FSPS algorithm and skipping to the step 9; 4, if the textures are horizontal textures, executing the step 7, if the textures are vertical textures, executing the step 5, if the textures are checkerboard textures, setting the integer pixel MV to be the optimal MV and skipping to the step 9, and otherwise, adopting the FSPS algorithm and skipping to the step 9; 5, interpolating in the vertical direction; 6, searching in the vertical direction, and skipping to the step 9; 7, interpolating in the horizontal direction; 8, searching in the horizontal direction; and 9, ending.
Description
Technical field
The invention belongs to the technical field of HEVC Video coding, relate to particularly a kind of based on HEVC'sMethod for estimating.
Background technology
HEVC (HighEfficiencyVideoCoding, efficient video coding) is up-to-date lookingFrequently coding standard. In HEVC coding standard, estimation can effectively be removed video sequence consecutive frameTime redundancy, has determined that coding rate, compression ratio and the decoding of video encoder looked to a great extentFrequently quality. Therefore, HEVC motion estimation module has increased Multi-encoding technology, for example, adopt fixing one dimension7 taps or the wave filter of 8 taps based on DCT produce respectively the brightness value of 1/2 and 1/4 pixel. HEVCPerformance surmounted all video encoders in the past, under the prerequisite of same-code quality, HEVC is relativeIn the code check that H.264 can save 50%. The high-performance of HEVC is the generation that rises to computation complexityValency.
In HEVC encoder, estimation comprises whole pixel and fraction pixel two parts. Fraction pixel motionEstimation is at the enterprising row interpolation in basis that obtains whole pixel motion vector (MV:motionvector)Computing, search obtains the process of fraction pixel precision MV. Fraction pixel estimation is in compressed video matterIn amount and compression ratio, can greatly improve the performance of encoder. But, due to extra computing, as insertedValue and fraction pixel are searched for, and fraction pixel estimation has greatly increased whole motion estimation moduleAmount of calculation.
The fraction pixel motion estimation algorithm that HEVC adopts is fraction pixel full-search algorithm (FSPS:FullSub-pixelSearch). As shown in Figure 1, I is the integer picture being obtained by integer pixel estimationElement position, h1, h2, h3 ..., h8 is I 8 half-pixel position around. FSPS algorithm calculatesThe absolute error of these 8 positions and (SAD:sumofabsolutedifference), and therefromSelect the position of SAD minimum. The SAD minimum of h2 in Fig. 1. Then, calculate optimum half-pix positionPut 8 1/4th location of pixels q1 around of h2, q2, q3 ... the SAD of q8, and therefrom selectOptimal location. The SAD minimum of q8, so q8 is final optimal location.
Fig. 2 is HEVC brightness layer fractional pixel interpolation schematic diagram. ai,j、bi,j、ci,j、di,j、hi,jAnd ni,jOnly just can be obtained by integer pixel interpolation. And to obtain ei,j、fi,j、gi,j、ii,j、ji,j、ki,j、pi,j、qi,jAnd ri,j, first to calculate ai,j,bi,jAnd ci,j, then in the vertical directionThey are carried out to interpolation, so just obtained all fraction pixels.
Estimation is the key technology of various video encoding standards. Along with the proposition of whole pixel algorithm withUpdate, can in the situation that only detecting 5 search points, find optimal solution. By contrast,The search finding amount of calculation of fraction pixel just seems and is difficult to accept, and therefore develops the quick of fraction precisionMotion estimation algorithm, has active demand.
Summary of the invention
Technology of the present invention is dealt with problems and is: overcome the deficiencies in the prior art, provide a kind of based on HEVCMethod for estimating, it can reduce the complexity of estimation under the prerequisite that does not reduce coding qualityDegree, thereby computation amount, computational speed improves greatly.
Technical solution of the present invention is: this method for estimating based on HEVC, comprises followingStep:
(1) current predicting unit PU is carried out to integer pixel estimation;
(2) judging that whether current PU fraction pixel estimation is invalid, is integer pixel MV to be putFor optimum MV and jump to step (9), otherwise execution step (3);
(3) judge whether current PU comprises textural characteristics, be to perform step (4), otherwise adoptFSPS algorithm also jumps to step (9);
(4) carry out the detection of texture classification, if horizontal texture performs step (7), if perpendicularStraight grain performs step (5), if gridiron pattern texture is set to integer pixel MVOptimum MV also jumps to step (9), if current PU does not have above textural characteristics,Adopt FSPS algorithm and jump to step (9);
(5) carry out vertical direction interpolation;
(6) carry out vertical direction search, jump to step (9);
(7) carry out horizontal direction interpolation;
(8) carry out horizontal direction search;
(9) finish.
One aspect of the present invention by judge integer pixel estimation effect enough good region skip invalidFraction pixel motion search, determine different fraction pixel searcher according to texture analysis on the other handMethod therefore reduces the complexity of estimation under the prerequisite that does not reduce coding quality, thus amount of calculationGreatly reduce, computational speed improves greatly.
Brief description of the drawings
Fig. 1 shows the fraction pixel motion estimation algorithm of HEVC in prior art.
Fig. 2 is HEVC brightness layer fraction pixel schematic diagram.
Fig. 3 a shows the fraction pixel searching algorithm based on vertical texture, and Fig. 3 b shows based on waterThe fraction pixel searching algorithm of flat grain.
Fig. 4 is according to the flow chart of the method for estimating based on HEVC of the present invention.
Detailed description of the invention
As shown in Figure 4, this method for estimating based on HEVC, comprises the following steps:
(1) current predicting unit PU is carried out to integer pixel estimation;
(2) judging that whether current PU fraction pixel estimation is invalid, is integer pixel MV to be putFor optimum MV and jump to step (9), otherwise execution step (3);
(3) judge whether current PU comprises textural characteristics, be to perform step (4), otherwise adoptFSPS (fraction pixel full-search algorithm, FullSub-pixelSearch) algorithm is also jumpedForward step (9) to;
(4) carry out the detection of texture classification, if horizontal texture performs step (7), if perpendicularStraight grain performs step (5), if gridiron pattern texture is set to integer pixel MVOptimum MV also jumps to step (9), if current PU does not have above textural characteristics,Adopt FSPS algorithm and jump to step (9);
(5) carry out vertical direction interpolation;
(6) carry out vertical direction search, jump to step (9);
(7) carry out horizontal direction interpolation;
(8) carry out vertical direction search;
(9) finish.
One aspect of the present invention by judge integer pixel motion search effect enough good region skip invalidFraction pixel motion search, determine different fraction pixel searcher according to texture analysis on the other handMethod therefore reduces the complexity of estimation under the prerequisite that does not reduce coding quality, thus amount of calculationGreatly reduce, computational speed improves greatly.
Preferably, in described step (2), decide current PU fraction pixel motion with formula (1)Estimate whether be invalid:
Wherein, MVIntBe integer pixel MV, T is MAE (MVInt) threshold value, MAE (MVInt) determineJustice is formula (2)
Wherein, SAD (MVInt) be absolute error that integer pixel MV is corresponding and, width is current PUWidth, height is the height of current PU, T obtains according to formula (3):
T=-0.014×qp2+1.095×qp-14.07(3)
Wherein, qp is quantization parameter.
Preferably, in described step (6), only search for the fraction pixel obtaining by vertical direction interpolationPosition.
Preferably, in described step (8), only search for the fraction pixel obtaining by horizontal direction interpolationPosition.
The present invention will be described in more detail below.
1. skip selectively invalid fraction pixel search
Fraction pixel estimation can improve compressed video quality and compression ratio, but fraction pixel MVSearch needs huge amount of calculation. If the least absolute error that mark MV search obtains and (MSAD:Minimumsumofabsolutedifference) equal the MSAD that whole pixel MV search obtains,Select whole pixel MV as final MV, it is invalid that fraction pixel MV search is regarded as. Otherwise, if pointThe MSAD that number MV search obtains is less than the MSAD that whole pixel MV search obtains, fraction pixel MV searchFor effectively, select fraction pixel MV as final MV. Front 30 frames to 6 standard video sequence carry outExperiment is that the shared ratio of whole pixel MV is in table 1 by final MV.
The ratio (30 frames, QP=27) that the whole pixel MV of table 1 is shared
From experimental result, on average have to exceed 60% MV and be positioned at whole location of pixels. In fact, rightIn some region of frame of video, fraction pixel search is to the raising of coding efficiency not obvious, for invalidSearch. If can predict enough region units accurately of these integer pixel Search Results, it is invalid to skipFraction pixel search, just can reduce unnecessary calculating.
Before fraction pixel estimation, judge the integer pixel Search Results of frame of video enough accuratelyPredicting unit (PU:predictionunit) is to skip the key of invalid fraction pixel motion search.
Can determine whether to skip point by mean absolute error corresponding to the integer pixel MV of current PUThe search of number pixel motion. Whether skip_fraction_flag skips invalid fraction pixel estimationMark. It is defined as
Wherein MVIntBe integer pixel MV, T is MAE (MVInt) threshold value. MAE (MVInt) determineJustice is
Wherein SAD (MVInt) be absolute error and the (SAD:sumofabsolute that integer pixel MV is correspondingDifference), width is the width of current PU, and height is the height of current PU. By testingRelease the definition of threshold value T:
T=-0.014×qp2+1.095×qp-14.07(3)
Wherein qp is quantization parameter (QP, QuantizationParameter).
2. the fraction pixel estimation based on grain direction
If current PU comprises vertical texture, so the integer pixel of current PU is carried out to horizontal directionInterpolation can cause video fuzzy. In like manner, for the PU that comprises horizontal texture, to its integer pixelThe interpolation of carrying out vertical direction also can cause video fuzzy. So, for there being horizontal or vertical direction lineThe PU of reason, we can skip some by with the texture fraction pixel position that in the other direction interpolation obtains.
First, analyze vertical texture. For vertical texture, only need search by vertical direction interpolationThe fraction pixel position obtaining. In Fig. 2, only has di,j、hi,jAnd ni,jBy vertical direction interpolationObtain. As shown in Fig. 3 (a), larger square I is obtained by integer pixel motion searchInteger pixel positions, 8 little squares are I 8 half-pixel position around. Only calculate A, B twoThe SAD of individual position, then finds a best position. Suppose the SAD minimum of A, so optimumHalf-pixel position A around two 1/4th location of pixels of vertical direction is 1,2. So countCalculate the SAD of 1,2 two 1/4th location of pixels. Finally select the position of sad value minimum.
Secondly, the PU with horizontal texture is discussed. For horizontal texture, only need search to pass through waterThe fraction pixel position that flat directional interpolation obtains. In Fig. 2, only has ai,j、bi,jAnd ci,jBy erectingStraight directional interpolation obtains. As shown in Fig. 3 (b), larger square I is moved by integer pixelThe integer pixel positions that search obtains, 8 little squares are I 8 half-pixel position around. Only meterThe SAD that calculates C, two positions of D, then finds a best position. Suppose the SAD minimum of C,So around optimum half-pixel position C, two 1/4th location of pixels of vertical direction are 1,2.So calculate the SAD of 1,2 two 1/4th location of pixels. Finally select the position of sad value minimumPut.
Finally, if current PU is gridiron pattern shape texture (existing horizontal texture has again vertical texture),Can directly skip fraction pixel estimation. Because in Fig. 2 except ai,j、bi,j、ci,j、di,j、hi,jAnd ni,jAll positions all by Horizontal interpolation and vertically interpolation jointly obtain. These two kinds of interpolationIn any one all can cause video fuzzy.
3. the main framework of motion estimation algorithm
Based on analysis above, the main framework of motion estimation algorithm is proposed, as shown in Figure 4. ToolBody, the estimation flow process of current PU is as described below. First, current PU is carried out to integer pictureElement estimation. Whether decide current PU fraction pixel estimation with formula (1) is invalid. If fraction pixel estimation is invalid to current PU, skip so fraction pixel estimation,And integer pixel MV is set to optimum MV. If fraction pixel estimation is effective, need to detectThe textural characteristics of current PU. If current PU comprises gridiron pattern shape texture, skip so fraction pixelEstimation, and integer pixel MV is set to optimum MV. For vertical texture, should adopt perpendicularStraight cutting value and vertically search. For horizontal texture, should adopt Horizontal interpolation and horizon scan. IfCurrent PU does not have above-mentioned textural characteristics, adopts so the FSPS algorithm of HEVC.
Such scheme is applied in HEVC encoder, and has obtained obvious effect.
Experimental result shows, fraction pixel fast motion estimation algorithm is than fraction pixel full-search algorithm(FSPS) in the situation that Y-PSNR (PSNR) has small reduction (0.01dB), bitrateOn average increased by 0.02%, total scramble time decreased average 24.56%, the whole estimation timeOn average save 40.86%.
Table 2
The above, be only preferred embodiment of the present invention, not the present invention done any pro formaRestriction, any simple modification that every foundation technical spirit of the present invention is done above embodiment, etc.With changing and modifying, all still belong to the protection domain of technical solution of the present invention.
Claims (4)
1. the method for estimating based on HEVC, is characterized in that: comprise the following steps:
(1) current predicting unit PU is carried out to integer pixel estimation;
(2) judging that whether current PU fraction pixel estimation is invalid, is integer pixel MV to be putFor optimum MV and jump to step (9), otherwise execution step (3);
(3) judge whether current PU comprises textural characteristics, be to perform step (4), otherwise adoptFSPS algorithm also jumps to step (9);
(4) carry out the detection of texture classification, if horizontal texture performs step (7), if perpendicularStraight grain performs step (5), if gridiron pattern texture is set to integer pixel MVOptimum MV also jumps to step (9), if current PU does not have above textural characteristics,Adopt FSPS algorithm and jump to step (9);
(5) carry out vertical direction interpolation;
(6) carry out vertical direction search, jump to step (9);
(7) carry out horizontal direction interpolation;
(8) carry out horizontal direction search;
(9) finish.
2. the method for estimating based on HEVC according to claim 1, is characterized in that: instituteState in step (2) and to current PU fraction pixel estimation whether to decide with formula (1)Invalid:
Wherein, MVIntBe integer pixel MV, T is MAE (MVInt) threshold value, MAE (MVInt) determineJustice is formula (2)
Wherein, SAD (MVInt) be absolute error that integer pixel MV is corresponding and, width is current PU
Width, height is the height of current PU, T obtains according to formula (3):
T=-0.014×qp2+1.095×qp-14.07(3)
Wherein, qp is quantization parameter.
3. the method for estimating based on HEVC according to claim 2, is characterized in that: instituteState and in step (5), only carry out vertical direction interpolation and obtain fraction pixel position, described step(6) in, only search for the fraction pixel position obtaining by vertical direction interpolation.
4. the method for estimating based on HEVC according to claim 3, is characterized in that: instituteState and in step (7), only carry out horizontal direction interpolation and obtain fraction pixel position, described step(8) in, only search for the fraction pixel position obtaining by horizontal direction interpolation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510994165.XA CN105611299B (en) | 2015-12-25 | 2015-12-25 | A kind of method for estimating based on HEVC |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510994165.XA CN105611299B (en) | 2015-12-25 | 2015-12-25 | A kind of method for estimating based on HEVC |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105611299A true CN105611299A (en) | 2016-05-25 |
CN105611299B CN105611299B (en) | 2018-11-23 |
Family
ID=55990796
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510994165.XA Active CN105611299B (en) | 2015-12-25 | 2015-12-25 | A kind of method for estimating based on HEVC |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105611299B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI744660B (en) * | 2018-06-29 | 2021-11-01 | 大陸商北京字節跳動網絡技術有限公司 | Mer improvements |
WO2022061613A1 (en) * | 2020-09-23 | 2022-03-31 | 深圳市大疆创新科技有限公司 | Video coding apparatus and method, and computer storage medium and mobile platform |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1688165A (en) * | 2005-06-09 | 2005-10-26 | 上海交通大学 | Fast motion assessment method based on object texture |
US6996180B2 (en) * | 2001-09-05 | 2006-02-07 | Intel Corporation | Fast half-pixel motion estimation using steepest descent |
CN101699865A (en) * | 2009-10-28 | 2010-04-28 | 山东大学 | Method for fast searching mass movement self-adapting sub pixel |
CN101795409A (en) * | 2010-03-03 | 2010-08-04 | 北京航空航天大学 | Content adaptive fractional pixel motion estimation method |
CN104811728A (en) * | 2015-04-23 | 2015-07-29 | 湖南大目信息科技有限公司 | Video content self-adaptation motion search method |
-
2015
- 2015-12-25 CN CN201510994165.XA patent/CN105611299B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6996180B2 (en) * | 2001-09-05 | 2006-02-07 | Intel Corporation | Fast half-pixel motion estimation using steepest descent |
CN1688165A (en) * | 2005-06-09 | 2005-10-26 | 上海交通大学 | Fast motion assessment method based on object texture |
CN101699865A (en) * | 2009-10-28 | 2010-04-28 | 山东大学 | Method for fast searching mass movement self-adapting sub pixel |
CN101795409A (en) * | 2010-03-03 | 2010-08-04 | 北京航空航天大学 | Content adaptive fractional pixel motion estimation method |
CN104811728A (en) * | 2015-04-23 | 2015-07-29 | 湖南大目信息科技有限公司 | Video content self-adaptation motion search method |
Non-Patent Citations (1)
Title |
---|
祝世平等: "《基于H.264的内容自适应分数像素运动估计算法》", 《光电子·激光》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI744660B (en) * | 2018-06-29 | 2021-11-01 | 大陸商北京字節跳動網絡技術有限公司 | Mer improvements |
US11470304B2 (en) | 2018-06-29 | 2022-10-11 | Beijing Bytedance Network Technology Co., Ltd. | Virtual merge candidates |
US11627308B2 (en) | 2018-06-29 | 2023-04-11 | Beijing Bytedance Network Technology Co., Ltd. | TMVP derivation |
WO2022061613A1 (en) * | 2020-09-23 | 2022-03-31 | 深圳市大疆创新科技有限公司 | Video coding apparatus and method, and computer storage medium and mobile platform |
Also Published As
Publication number | Publication date |
---|---|
CN105611299B (en) | 2018-11-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101385356B (en) | Process for coding images using intra prediction mode | |
CN103260031B (en) | Method and apparatus for encoding/decoding to motion vector | |
CN103248895B (en) | A kind of quick mode method of estimation for HEVC intraframe coding | |
EP1734769B1 (en) | An encoder and encoding method for motion estimation | |
CN110087087A (en) | VVC interframe encode unit prediction mode shifts to an earlier date decision and block divides and shifts to an earlier date terminating method | |
CN103517069A (en) | HEVC intra-frame prediction quick mode selection method based on texture analysis | |
CN102263951B (en) | Quick fractal video compression and decompression method | |
CN100591136C (en) | Video frequency intraframe coding method based on null field decomposition | |
CN103327327B (en) | For the inter prediction encoding unit selection method of high-performance video coding HEVC | |
CN101860748A (en) | Side information generating system and method based on distribution type video encoding | |
CN102598670A (en) | Method and apparatus for encoding/decoding image with reference to a plurality of frames | |
CN106604035B (en) | A method of the estimation for Video coding and compression | |
CN101621694B (en) | Motion estimation method, motion estimation system and display terminal | |
CN105898332B (en) | For the fast deep figure frame mode decision method of 3D-HEVC coding standards | |
CN103188496A (en) | Fast motion estimation video encoding method based on motion vector distribution forecast | |
CN107623850A (en) | A kind of quick screen contents encoding method based on temporal correlation | |
CN101431675A (en) | Image element motion estimating method and apparatus | |
CN104349167B (en) | A kind of method of adjustment of Video coding rate-distortion optimization | |
CN102752588A (en) | Video encoding and decoding method using space zoom prediction | |
CN103237222A (en) | Motion estimation method adopting multi-mode searching manner | |
CN101299818B (en) | N level sub-pixel search method based on whole pixel searching result | |
CN110519591A (en) | A kind of prediction mode fast selecting method based on intraframe coding in multipurpose coding | |
CN104702959B (en) | A kind of intra-frame prediction method and system of Video coding | |
CN109120940B (en) | Adaptive factor video scaling motion estimation method | |
CN101783956B (en) | Backward-prediction method based on spatio-temporal neighbor information |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |