CN1558683A - Variable shape searching (VSS) quick motion estimation method - Google Patents
Variable shape searching (VSS) quick motion estimation method Download PDFInfo
- Publication number
- CN1558683A CN1558683A CNA2004100140484A CN200410014048A CN1558683A CN 1558683 A CN1558683 A CN 1558683A CN A2004100140484 A CNA2004100140484 A CN A2004100140484A CN 200410014048 A CN200410014048 A CN 200410014048A CN 1558683 A CN1558683 A CN 1558683A
- Authority
- CN
- China
- Prior art keywords
- search
- vss
- algorithm
- point
- rhombus
- 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.)
- Pending
Links
Images
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a search shape variable rapid motion estimation algorithm based on the combination search of diamond and hexagon which comprises, searching the top dot and the five central dots for the big diamond, performing concentrating search with the small diamonds, or finishing the residual search procedures with the mode 1 or mode 2 hexagon searches.
Description
Technical field
The present invention relates to a kind of fast motion estimation algorithm of the search shape variable (VSS) based on rhombus and hexagon Syndicating search, can be used for H.261, H.263, MPEG2, MPEG4, H.264 in the video coding of various sizes piece.
Background technology
Motion compensated predictive coding is the representative art that reduces the video image redundancy, and international standards such as MPEG-2, H263, MPEG-4 and H264 have all extensively adopted the motion compensated prediction technology.Stipulated method for video coding and code stream frame structure though it should be noted that a series of international standards, the specific algorithm of estimation and control strategy are open to the designer.BMA (BMA) is present the most frequently used motion estimation techniques, and it realizes that effect depends primarily on three factors: matching criterior (match criteria), hunting zone (search area) and way of search (search method).Practical matching criterior adopts mean absolute error (MAD) criterion mostly, and the hunting zone is a unit with macro block and piece generally, and way of search is the principal element that influences the BMA performance, also is the key technology in the motion estimation algorithm.In all BMAs, full search (FS) algorithm performance is best, but adopt the full search method of global optimum, estimation will account for the 60%-80% of whole coding operand, requiring to be difficult to practicality under the situation of real-time coding, therefore must between estimation precision and amount of calculation, compromise, how fast and effeciently carry out the major issue that estimation becomes the video coding technique that comprises hierarchical coding.
Existing various fast motion estimation algorithm all is according to the center-biased distribution character of video image motion vector field and the high correlation between vector, different motion content according to piece is determined its search starting point, hunting zone and search strategy, thereby the realization block motion vector is fast and effeciently estimated.Wherein, classical three-step approach (TSS) is because effectively simple, recommended by H.261 RM8 and the SM3 of MPEG.New three-step approach (NTSS) is improved on the basis of TSS and is formed, and the method that adopted for half step stopped has effectively reduced computing time.The corresponding largest motion displacement of four step rule (4SS) is ± 7 situation, and employing is the window that 9 interior test points of 5 * 5 windows at center replace 9 * 9 among the TSS with the datum mark.At present, rhombus algorithm (DS) is a kind of fast algorithm all relatively more outstanding on time performance and precision index, and its search shape is a rhombus.
Summary of the invention
Therefore, technical problem to be solved by this invention is: how fast and effeciently to realize estimation and motion compensation in the video coding, promptly for little, in, the image encoding of big motion, effectively reduce the amount of calculation of piece coupling, obtain reconstructed image quality and the signal to noise ratio close with other fast algorithm, and algorithm is simply effective, is convenient to software and hardware and realizes.
Technical scheme of the present invention is as follows:
Existing various fast algorithm all is according to the center-biased distribution character of video image motion vector and the high correlation between vector, different motion content according to piece is determined its search starting point, hunting zone and search strategy, thereby the realization block motion vector is fast and effeciently estimated.
Circular search pattern is the most effective beyond doubt for the motion vector of isotropic distribution.Yet a large amount of statistical experiments show that the likelihood ratio that general object moves in the horizontal and vertical directions is bigger, and the frequency spectrum of image becomes the rhomboid distribution of class more.Can prove that the regular polygon that can comprehensively cover a search plane has three kinds on equilateral triangle, square, regular hexagon, wherein, orthohexagonal Center Gap maximum in the inscribed polygon of circle, area coverage are also maximum.Because diamond search and hexagon search respectively possess some good points, in order to take all factors into consideration the advantage of these two kinds of search patterns better, we have proposed a kind of fast algorithm based on rhombus and orthohexagonal search shape variable, its basic thought is to adopt rhombus to carry out fine search, and regular hexagon then is used to carry out coarse search.
Algorithm of the present invention is better than popular several fast motion-estimation algorithms at present, not only greatly reduces the amount of calculation of piece coupling, and can obtain reconstructed image quality and the signal to noise ratio close with other fast algorithm.For little, in, the coding of big moving image, this algorithm all has performance improvement, particularly in the estimation effect of big motion vector improve more obvious.This algorithm is simply effective, is convenient to software and hardware and realizes that versatility is stronger, can apply at present widely popular H.261, H.263, MPEG-2, MPEG-4, H.264 wait in the video coding international standard, improve the performance of all kinds of coding and decoding videos thus.
Description of drawings
Fig. 1 is two kinds of basic search patterns of VSS algorithm, and Fig. 1 wherein-(a) is a regular hexagon horizontal direction search pattern schematic diagram, and Fig. 1 (b) is a regular hexagon vertical direction search pattern schematic diagram
Fig. 2 is a VSS algorithm search flow chart
Fig. 3 is a VSS algorithm search step example
1. among Fig. 1 is big rhombus; 2. be regular hexagon; 3. be little rhombus
Among Fig. 21 ' is step 1: calculate that to be positioned at search window center and step-length be 2 big rhombus summit and supercentral 5 points; Is 2 ' that the minimum MAD point is positioned at the center? 3 ' is step 2: the deterministic model type, and calculate remaining 4 of this pattern regular hexagon and do not search for candidate point; 4 ' is step 3: the regular hexagon minimum MAD summit of choosing with back is formed centrally new regular hexagon in being, calculates 3 new candidate points; Is 5 ' that the minimum MAD point is positioned at the center? 6 ' is step 4: it is 1 little rhombus that the minimum MAD point of choosing with back is formed centrally step-length in being, carries out final step and focuses on fine search, and the position at minimum MAD point place is exactly the final goal motion vector in little rhombus summit and supercentral 5 points.
Embodiment
In order to understand characteristics of the present invention, function and effect better, the following example of existing utilization is elaborated also in conjunction with the accompanying drawings.
This algorithm has two kinds of basic search patterns (as Fig. 1): pattern 1 and pattern 2, shape is asymmetric in the horizontal and vertical directions to consider regular hexagon, and the probability that motion vector occurs on this both direction is relatively large, so two kinds of patterns adopt the regular hexagon with different directions respectively.Regular hexagon among Fig. 1 is made up of 7 points that are labeled as " 1. ", and the basic search step-length is 2, lacks 2 points than 9 diamond search pattern, and one of them point is positioned at the center, and other six points are distributed on the orthohexagonal summit.Big rhombus is made up of four points that are labeled as " 2. ", and step-size in search is 2, and preference pattern type when it is used for initial search wherein has two points to overlap with point on the regular hexagon.Little rhombus is made up of four points that are labeled as " 3. ", and step-size in search is 1, and it is mainly used in the focusing fine search of estimation final stage.In the process of search, the center of regular hexagon region of search can move on corresponding six summits.No matter which summit is regular hexagon integral body move on, always occurs three new not search summits, the then searched mistake in other three summits in the new regular hexagon.Corresponding, diamond search is each when mobile in the DS algorithm, 3 or 5 new not search points will occur along different directions, and about 4 new not search points will appear in average each whole moving, and has more a point than the method for author's proposition.
Consider the distribution character and the orthohexagonal true form of motion vector, this algorithm first step is at first carried out the selection of search pattern.Elder generation's step-size in search is 2 big rhombus summit and five points at center.If the minimum MAD point is positioned at big rhombus center, just adopting step-length is that 1 little rhombus focuses on fine search, and this need search for four points on the little rhombus summit again, and wherein the position of minimum MAD point is exactly a target motion vectors; If the minimum MAD point is the summit on the big rhombus horizontal direction, adopt " pattern 1 " to finish remaining search, otherwise, if the summit of big rhombus vertical direction just adopts " pattern 2 " to finish remaining search.Two kinds of patterns all will adopt fixing separately regular hexagon to search for, and each regular hexagon integral body all has only three new optimal candidate points to need to calculate when moving on the new summit.So repeatedly, when the MAD smallest point is positioned at orthohexagonal center, shows and extreme point to occur, focus on fine search so forward final step to.If regular hexagon shifts out the hunting zone, also will carry out the final step search, at this moment need to carry out border extended, optimum movement vector will be chosen the minimum MAD point in the hunting zone.Fig. 2 is the general search step flow chart of VSS algorithm.
To further set forth the VSS basic idea by the object lesson among Fig. 3 below.At first calculating apart from search window center step-length is 2 the big rhombus summit and the MAD value of supercentral 5 points (being labeled as the point of " 1 " among Fig. 3), comparative result minimum MAD point is the summit on the big rhombus horizontal direction, therefore adopts " pattern 1 " to finish remaining search procedure.Then calculate and relatively remaining 4 of " pattern 1 " regular hexagon do not search for candidate point (being labeled as the point of " 2 " among Fig. 3), the minimum MAD point is the summit in the regular hexagon lower right corner as a result; Be formed centrally new regular hexagon in this summit being then, calculate 3 new candidate points (being labeled as the point of " 3 " among Fig. 3), if minimum MAD point is still on orthohexagonal summit, the regular hexagon minimum MAD summit of choosing with back is formed centrally new regular hexagon in being, calculate 3 new candidate points (being labeled as the point of " 4 " and " 5 " among Fig. 3), with this repeatedly, at last to be formed centrally step-length in being be 1 little rhombus to the minimum MAD point of choosing with the 5th step, carry out the focusing fine search of final step, the position (+4 ,+3) at minimum MAD point place is exactly the final goal motion vector that we will look on little rhombus summit and in supercentral 5 points (being labeled as the point of " 6 " among Fig. 3).VSS algorithm among Fig. 3 has finally searched target motion vectors (+4 ,+3) by 6 steps, needs the candidate target point of search to have 22.Certainly, just for the thought of VSS algorithm is described, do not need so big amount of calculation under in fact a lot of situations for this example.
Claims (9)
1. fast motion estimation algorithm one VSS who searches for shape variable, this algorithm mainly comprises the steps:
A. at first determine search starting point and step-length, utilize big rhombus to carry out the selection of search pattern;
B. adopt regular hexagon then to carry out rough search;
C. on this basis, adopt little rhombus to carry out fine search at last.
2. VSS algorithm according to claim 1 is characterized in that, the search starting point can be predicted value point or origination data, and initial step length can be 2 or 1 according to the match block size with the hunting zone, and matching criterior adopts mean absolute error (MAD) criterion.
3. VSS algorithm according to claim 1, it is characterized in that, shape is asymmetric in the horizontal and vertical directions to consider the uneven distribution characteristic of actual motion vector and regular hexagon, and the VSS algorithm at first carries out the search of big rhombus, carries out the selection of search pattern then.
4. VSS algorithm according to claim 1 is characterized in that, absorbs diamond search and hexagon search strong point separately, and these two kinds of ways of search of integrated use adaptively, this method can apply in the block matching motion estimation of all size.
5. VSS algorithm according to claim 3, it is characterized in that, search for five points at big rhombus summit and center earlier, if the minimum MAD point is positioned at big rhombus center, just adopt little rhombus to carry out focused search, wherein the position of minimum MAD point is exactly a target motion vectors; If the minimum MAD point is the summit on the big rhombus horizontal direction, adopt " pattern 1 " hexagon to finish remaining search procedure, otherwise, if the summit of big rhombus vertical direction just adopts " pattern 2 " hexagon to finish remaining search procedure.
6. VSS algorithm according to claim 4, it is characterized in that, the regular hexagon minimum MAD summit of choosing with back is formed centrally new regular hexagon in being, adopt regular hexagon then to carry out coarse search, when moving on the new summit, each regular hexagon integral body all have only three new optimal candidate points to need to calculate, so repeatedly, and when the MAD smallest point is positioned at orthohexagonal center, show and extreme point to occur, adopt little rhombus to carry out focused search so forward final step to.
7. VSS algorithm according to claim 6 is characterized in that, the searching times that each piece coupling needs in the VSS algorithm can come out by formula form quantitative expression.
8. VSS algorithm according to claim 6, it is characterized in that, the VSS algorithm can be specially adapted to the coding of big sport video, and reconstructed image quality and other fast algorithm be suitable substantially to search same target motion vectors than DS algorithm candidate point still less.
9. VSS algorithm according to claim 7 is characterized in that VSS algorithm thinking is simple, is convenient to software and hardware and realizes.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNA2004100140484A CN1558683A (en) | 2004-02-13 | 2004-02-13 | Variable shape searching (VSS) quick motion estimation method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNA2004100140484A CN1558683A (en) | 2004-02-13 | 2004-02-13 | Variable shape searching (VSS) quick motion estimation method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN1558683A true CN1558683A (en) | 2004-12-29 |
Family
ID=34351257
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CNA2004100140484A Pending CN1558683A (en) | 2004-02-13 | 2004-02-13 | Variable shape searching (VSS) quick motion estimation method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN1558683A (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN100409692C (en) * | 2005-01-31 | 2008-08-06 | 凌阳科技股份有限公司 | Motion-vector micro-searching mode determining system and method |
CN1941909B (en) * | 2005-09-28 | 2010-05-05 | 中国科学院自动化研究所 | Fast motion evaluation based on orthogonal distributing model |
CN101621694B (en) * | 2009-07-29 | 2012-01-11 | 深圳市九洲电器有限公司 | Motion estimation method, motion estimation system and display terminal |
CN102790883A (en) * | 2012-07-26 | 2012-11-21 | 中国航天科工集团第三研究院第八三五七研究所 | Image compression movement vector searching method |
CN103024390A (en) * | 2012-12-21 | 2013-04-03 | 天津大学 | Self-adapting searching method for motion estimation in video coding |
CN106101722A (en) * | 2016-06-07 | 2016-11-09 | 成都金本华电子有限公司 | The quasi-all direction search method of layering improved based on YANG structure and system |
CN106650551A (en) * | 2015-10-30 | 2017-05-10 | 国网山西省电力公司电力科学研究院 | High-voltage breaker moving contact movement track real-time identification method based on priori knowledge |
-
2004
- 2004-02-13 CN CNA2004100140484A patent/CN1558683A/en active Pending
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN100409692C (en) * | 2005-01-31 | 2008-08-06 | 凌阳科技股份有限公司 | Motion-vector micro-searching mode determining system and method |
CN1941909B (en) * | 2005-09-28 | 2010-05-05 | 中国科学院自动化研究所 | Fast motion evaluation based on orthogonal distributing model |
CN101621694B (en) * | 2009-07-29 | 2012-01-11 | 深圳市九洲电器有限公司 | Motion estimation method, motion estimation system and display terminal |
CN102790883A (en) * | 2012-07-26 | 2012-11-21 | 中国航天科工集团第三研究院第八三五七研究所 | Image compression movement vector searching method |
CN102790883B (en) * | 2012-07-26 | 2015-06-10 | 中国航天科工集团第三研究院第八三五七研究所 | Image compression movement vector searching method |
CN103024390A (en) * | 2012-12-21 | 2013-04-03 | 天津大学 | Self-adapting searching method for motion estimation in video coding |
CN103024390B (en) * | 2012-12-21 | 2015-09-09 | 天津大学 | For the self-adapted search method of the estimation in Video coding |
CN106650551A (en) * | 2015-10-30 | 2017-05-10 | 国网山西省电力公司电力科学研究院 | High-voltage breaker moving contact movement track real-time identification method based on priori knowledge |
CN106101722A (en) * | 2016-06-07 | 2016-11-09 | 成都金本华电子有限公司 | The quasi-all direction search method of layering improved based on YANG structure and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
RU2597521C1 (en) | Method and apparatus for encoding/decoding motion vector | |
CN108134939B (en) | Motion estimation method and device | |
CN108419082B (en) | Motion estimation method and device | |
Li et al. | Fast multi-frame motion estimation algorithm with adaptive search strategies in H. 264 | |
KR100994768B1 (en) | Motion estimation method for encoding motion image, and recording medium storing a program to implement thereof | |
CN103237222A (en) | Motion estimation method adopting multi-mode searching manner | |
CN100551071C (en) | Integer pixel quick mixing search method based on the center prediction | |
Tsai et al. | A novel 3-D predict hexagon search algorithm for fast block motion estimation on H. 264 video coding | |
CN102647598A (en) | H.264 inter-frame mode optimization method based on maximin MV (Music Video) difference value | |
CN1558683A (en) | Variable shape searching (VSS) quick motion estimation method | |
CN106658024B (en) | A kind of quick method for video coding | |
Liu et al. | Research and Optimization of Low-Complexity Motion Estimation Method Based on Visual Perception. | |
WO2012010023A1 (en) | Method and apparatus for image motion estimation | |
Aoun et al. | Multiresolution motion estimation and compensation for video coding | |
WO2022218299A1 (en) | Motion estimation method and apparatus, electronic device, and readable storage medium | |
JP2012129791A (en) | Image encoder | |
Ahmed et al. | Mean Predictive Block Matching (MPBM) for fast block-matching motion estimation | |
KR20040106202A (en) | Method and apparatus for motion vector search | |
Olivieri et al. | Noise-robust recursive motion estimation for H. 263-based videoconferencing systems | |
Babu et al. | Performance analysis of block matching algorithms for highly scalable video compression | |
CN105611299A (en) | Motion estimation method based on HEVC | |
Fazli et al. | A Fast Adaptive Rood Pattern Algorithm for Video Compression | |
CN117412065B (en) | Optimization scheme of spiral search algorithm | |
Jiang et al. | Low complexity variable block size motion estimation algorithm for video telephony communication | |
Liu et al. | A self-adaptive and fast motion estimation search method for H. 264/AVC |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |