CN1741418A - Adaptive reference frame selecting method based on mode inheritance in multiframe movement estimation - Google Patents
Adaptive reference frame selecting method based on mode inheritance in multiframe movement estimation Download PDFInfo
- Publication number
- CN1741418A CN1741418A CN200510094355.2A CN200510094355A CN1741418A CN 1741418 A CN1741418 A CN 1741418A CN 200510094355 A CN200510094355 A CN 200510094355A CN 1741418 A CN1741418 A CN 1741418A
- Authority
- CN
- China
- Prior art keywords
- patterns
- reference frame
- estimation
- mode
- frames
- 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
- 238000000034 method Methods 0.000 title claims abstract description 20
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 6
- 230000015572 biosynthetic process Effects 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 2
- 238000013517 stratification Methods 0.000 abstract 2
- 238000012360 testing method Methods 0.000 description 7
- 230000006835 compression Effects 0.000 description 6
- 238000007906 compression Methods 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
Images
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
A method for selecting adaptive reference frame in multiple frame moving estimation based on mode inheritance includes using a new reference queue formed by moving estimation result of high stratification mode to carry out moving estimation for low stratification mode , setting up reference frame optimum seeking list of relevant mode i.e. 16 x 16 mode first , then 16 x 8 mode and finally 8 x 8 mode as well as its submode according to specific inheritance relation of each mode before moving estimation is carried out .
Description
One, technical field
The present invention relates to the algorithm of the interframe encode estimation in the video compression coding (H.264/MPEG-4AVC).
Two, background technology
H.264/MPEG-4AVC be up-to-date video encoding standard by JVT work exploitation, one of its advantage is the compression ratio height, with respect to H.263, it has saved 50% code check ([2] B.Girod and M.Flierl, " Multi-framemotion-compensated video compression for the digital set-top box (many pictures sport video compression compensation of set-top box) ", Proc.IEEE ICIP, Sept.2002).This standard allows use to reach 16 reference frames and seven kinds of patterns are used for estimation, has improved the precision of prediction of interframe encode significantly.But the improvement of its compression efficiency is a cost to increase amount of calculation and complexity, wherein the computational complexity of multiframe estimation and operand increase with the reference frame quantity of using is linear, 80% computing capability that we can say encoder all expends ([3] P.Pirsch on the estimation in interframe encode, N.Demassieux, and W.Gehrke, " VLSI architectures for video compression-a survey ", Proc.IEEE, vol.83, no.2, pp.220-246, Feb 1995).The increase of operand and computational complexity has further increased the realization difficulty of encoder.This is applied to codec in the hardware for need, is a challenge when especially having only on the mobile computing device of finite computational abilities.And under different patterns, each reference frame obtainable distortion gain be correlated with.Therefore, in seven kinds of patterns all candidate frame all being scanned not is high-efficiency method.
Three, summary of the invention
The present invention seeks to: propose a new reference frame method for optimizing, be used for reducing the operand of multiframe motion estimation process, and can guarantee RD (distortion rate) performance much at one based on succession.Especially H.264 the adaptive reference frame selecting method in the multiframe estimation based on mode inheritance.
The object of the present invention is achieved like this: its basic skills is to use a new reference formation when the pattern of lower level is carried out estimation (ME), the motion estimation result of the construction basis higher level pattern of this formation, for example the reference formation of 8 * 8 patterns is with reference to the result of 16 * 16 patterns.Reduce with interval time and the principle that strengthens according to two two field picture correlations, each candidate's reference frame should be taked different weights in estimation; For each pattern, only need carry out motion estimation operation to the several reference frames in front, just can guarantee that certain predictor is right; Secondly, the reference frame selection of following layer model can be considered the inheritance that estimation predicts the outcome, and promptly the estimation based on upper mode predicts the outcome, and selects in the upper mode estimation more excellent several reference frames that predict the outcome to carry out estimation; At every kind of inheritance that pattern is concrete, before carrying out estimation, set up the reference frame preference lists of corresponding modes in the algorithm.
Specific implementation process of the present invention is as follows: when carrying out the ME of each macro block, the order of model selection is to carry out according to descending order, i.e. 16 * 16 patterns at first, and 16 * 8 patterns, 8 * 16 patterns are 8 * 8 patterns and its subpattern at last then.For 16 * 16 patterns, all candidate's reference frames are carried out estimation.For 16 * 8 patterns and 8 * 16 patterns, new reference frame lists comprises reference frame 1,2,3; Three reference frames that also comprise RD expense minimum after 16 * 16 pattern estimation.For 8 * 8 patterns, because its motion estimation result need be as the foundation of 8 * 4,4 * 8 and 4 * 4 schema reference frames tabulation foundation, therefore the reference frame lists that is used for 8 * 8 pattern estimation has also comprised reference frame 4 and 5 with respect to 16 * 8 patterns and 8 * 16 patterns.The reference frame lists of 8 * 4 patterns and 4 * 8 patterns set up more complicated.In the tabulation, except that comprising reference frame 1,2,3, also comprise three reference frames of the RD expense minimum of 8 * 8 affiliated model predictions, comprise three reference frames of the RD expense minimum of 16 * 8 or 8 * 16 model predictions that it is affiliated in addition.The reference frame lists of 4 * 4 patterns comprises three reference frames of the RD expense minimum of reference frame 1,2,3 and 8 * 8 model predictions.
In the methods of the invention, only need candidate's reference frame of search part, rather than check each macro block of each reference frame.The selection of the reference frame of searching for is determined according to the result of the ME that has carried out.By the probability distribution to ME (estimation) expense minimum value, and the correlation between the ME expense minimum value is analyzed between the different mode, and its result has impelled the foundation based on the reference frame selecting method of inheriting.
Mechanism of the present invention: in JM prototype software H.264, it is exactly to seek to use which reference frame under which kind of pattern that the RD of estimation optimizes, and makes its RD expense minimum.Its implementation procedure is as follows: at 7 kinds of inter-frame forecast modes 16 * 16,16 * 8,8 * 16,8 * 8,8 * 4,4 * 8 and 4 * 4 and intra prediction mode, all reference frames are carried out estimation.As reference frame quantity is 16, then need carry out 16 * 7=112 time estimation to each macro block, and computing cost is very big.Yet each reference frame is for the same meaning of encoding compression, and the result's of the macro block of same position or piece estimation correlation between different mode all needs us to consider simultaneously.
Common two two field picture correlations reduce with interval time and strengthen, that is to say, in 16 reference frames, under 7 kinds of different modes, the position is 1,2,3 reference frame (being three nearest frames), and the reference frame after the possibility of RD expense minimum is leaned on than the position is greater.The existence of this rule of verification experimental verification.Table one is 6 kinds of image sequences when carrying out ME, under the same prediction mode, and the probability distribution of different reference frame RD expense minimums.As seen from Table 1, the probability of nearest three frame RD expense minimums on average surpasses 75%.
The reference frame probability distribution of table 1, expense minimum
| Foreman | mother& daughter | Hall_monitor | News | carphone | container | On average | |
| Reference frame 1 | 52.23 | 64.55 | 67.27 | 77.15 | 46.22 | 68.3 | |
| | 13.37 | 8.76 | 6.18 | 6 | 12.66 | 6.75 | |
| | 7.96 | 4.7 | 3.39 | 2.72 | 8.24 | 3.42 | |
| Amount to | 73.56 | 78.01 | 76.84 | 85.87 | 67.12 | 78.47 | 76.645 |
The degree of conformity of reference frame ME minimum under table 2, the different ME pattern
| With reference to the ME pattern | The ME pattern | Foreman | Mother& daughter | hall_ monitor | news | Carphone | container | On average |
| 16×16 | 16×8 | 67.13 | 75.26 | 80.86 | 85.19 | 64.59 | 77.59 | |
| 16×16 | 8×16 | 65.32 | 73.12 | 80.49 | 82.91 | 63.24 | 78.61 | |
| 16×16 | 8×8 | 54.24 | 67.45 | 77.67 | 80.04 | 51.74 | 72.9 | |
| 8×8 | 8×4 | 83.7 | 83.98 | 89.34 | 88.29 | 81.37 | 90.09 | |
| 8×8 | 4×8 | 77.78 | 83.58 | 91.85 | 88.59 | 78.47 | 89.76 | |
| 8×8 | 4×4 | 74.15 | 80.51 | 90.41 | 86.73 | 74.12 | 87.91 | |
| 8×8 | On average | 70.39 | 77.32 | 85.10 | 85.29 | 68.92 | 82.81 | 78.31 |
Secondly, when carrying out the ME of macro block, the order of model selection is to carry out according to descending order, i.e. 16 * 16 patterns at first, and 16 * 8 patterns, 8 * 16 patterns are 8 * 8 patterns and its subpattern at last then.Because a part of the pattern that smaller pattern is normally bigger, therefore the pattern ME result who carries out earlier has directive significance for the ME that carries out thereafter.For example in the predicting the outcome of 16 * 16 patterns, the reference frame of its expense minimum is also to be minimum probably carrying out 8 * 8 pattern ME.By test data analyzer, this hypothesis is verified.Table 2 is that 6 kinds of cycle testss are when carrying out ME, the identical probability distribution of reference frame of RD expense minimum between the different mode, as when handling the Foreman sequence, 16 * 8 probabilities identical with the reference frame of expense minimum in 16 * 16 two-modes are 67.13%, and 8 * 4 probabilities identical with the reference frame of expense minimum in 8 * 8 two-modes are 89.34% when handling the hall-monitor sequence.
Based on above-mentioned analysis, can find that each candidate's reference frame should take different weights in ME.For each pattern, only need carry out the ME operation to the several reference frames in front, just can guarantee that certain predictor is right.Secondly, the reference frame selection of following layer model can be considered the inheritance that ME predicts the outcome, and promptly the ME based on upper mode predicts the outcome, and selects in the upper mode ME more excellent several reference frames that predict the outcome to carry out ME.Just be based on this thought based on the reference frame selection algorithm of inheriting.At every kind of inheritance that pattern is concrete, before carrying out estimation, set up the reference frame preference lists of corresponding modes in the algorithm.
For 16 * 16 patterns, because estimation, should be carried out to all candidate's reference frames in the basis that the new tabulations that are all of its motion estimation result are set up therefore.For 16 * 8 patterns and 8 * 16 patterns, new reference frame lists comprises reference frame 1,2,3; Three reference frames that also comprise RD expense minimum after 16 * 16 pattern estimation.For 8 * 8 patterns, because its motion estimation result need be as the foundation of 8 * 4,4 * 8 and 4 * 4 schema reference frames tabulation foundation, therefore the reference frame lists that is used for 8 * 8 pattern estimation has also comprised reference frame 4 and 5 with respect to 16 * 8 patterns and 8 * 16 patterns.The reference frame lists of 8 * 4 patterns and 4 * 8 patterns set up more complicated.In the tabulation, except that comprising reference frame 1,2,3, also comprise three reference frames of the RD expense minimum of 8 * 8 affiliated model predictions, comprise three reference frames of the RD expense minimum of 16 * 8 or 8 * 16 model predictions that it is affiliated in addition.When 8 * 4 and 4 * 8 schema reference frames are preferred the ME of 16 * 8 and 8 * 16 patterns of correspondence as a result references object referring to Fig. 1, be the ME result of the preferred reference block A of reference frame (16 * 8 patterns or 8 * 16 patterns) of piece 1-4 among Fig. 1 (8 * 4 patterns or 4 * 8 patterns), i.e. the ME result of the preferred reference block B of reference frame (16 * 8 patterns or 8 * 16 patterns) of piece 5-8 (8 * 4 patterns or 4 * 8 patterns) among the figure.The reference frame lists of 4 * 4 patterns comprises three reference frames of the RD expense minimum of reference frame 1,2,3 and 8 * 8 model predictions.The realization flow of specific algorithm is referring to Fig. 2.The realization details of the inventive method is stated as follows, comprises the result of analysis mode test.
The present invention proposes that a simple effective method is used to reduce because the computing cost that causes of multiframe estimation, the RD performance does not obviously descend simultaneously.Result of the test shows that this method guarantees RD (distortion rate) performance much at one simultaneously with respect to saving about 50% operand usually under multi-reference frame (16 frame) situation.
Four, description of drawings
Fig. 1 is the present invention 8 * 4 (Figure 1A) and the preferred references object of 4 * 8 (Figure 1B) schema reference frame
Fig. 2 is the flow chart that the present invention is based on the reference frame optimization algorithm realization of succession
Fig. 3 be traditional JM algorithm of the present invention's 5 reference frames and 10 reference frames new algorithm the RD curve ratio
Fig. 4 be traditional JM algorithm of the present invention's 10 reference frames and 16 reference frames new algorithm the RD curve ratio
Five, embodiment
As shown in Figure 1, 2, the estimation of 16 * 16 patterns (ME) result is the basis that all new tabulations are set up, and therefore, should carry out estimation to all candidate's reference frames.For 16 * 8 patterns and 8 * 16 patterns, new reference frame lists comprises reference frame 1,2,3 on this basis; Three reference frames that also comprise RD expense minimum after 16 * 16 pattern estimation.Again to 8 * 8 pattern estimation.8 * 8 pattern motion estimation result foundation that tabulation is set up as 8 * 4,4 * 8 and 4 * 4 schema reference frames again, therefore the reference frame lists that is used for 8 * 8 pattern estimation has also comprised reference frame 4 and 5 with respect to 16 * 8 patterns and 8 * 16 patterns.The reference frame lists of 8 * 4 patterns and 4 * 8 patterns set up more complicated.As Fig. 2 and above-mentioned.QP: the i.e. default quantization parameter that the DCT discrete cosine transform coefficient is quantized.
Algorithm of the present invention correct in JM8.6 realizes.To Forman, Mother﹠amp; Daughter, Hall Monitor, News, Carphone and six QCIF of Container (cycle tests) (1/4 CLV Common Intermediate Format) sequence use JM8.6 (calculating the interpolation of all sub-pixel points, in the method for motion compensation and sub-pixel search) and the amended software before revising to carry out encoded test respectively.Coding range is frame 0-299 totally 300 frames, sees for simply opening, and adopts full P frame coding, and the region of search is 16, uses fast motion estimation (UseFME=1).Be operand and RD performance relatively, when being 5,10 and 16 and in the scope of QP from 25 to 34, use two kinds of above-mentioned methods to encode at reference frame respectively.Table 3 is estimation time (METime) and RD (PSNR of average bit rate and the brightness value) comparative results behind the coding.Estimation time in the table and RD are the mean value the QP from 25 to 34.Wherein SNRY represents the Y-PSNR of Y component.By table 3 as seen, the new algorithm that adopts 10 reference frames is with respect to the traditional JM algorithm that adopts 5 reference frames, and ME saves above 10% operation time, and the RD performance improves code check decline 1% when promptly mean P SNR has improved 0.03db.50% of the traditional JM algorithm that adopts the ME of the new algorithm of 16 reference frames to have only operation time to adopt 10 reference frames, but mean P SNR do not reduce, and code check also descends to some extent, promptly the RD performance also obtains part and improves.
The RD performance of table 3, two pairs of different coding modes and ME compare operation time
| Reference frame number ﹠ algorithm types | Forema n | Mother& daughte r | hall_ Monito | news | Carphon e | contain er | On average | |
| SNRY 5&JM8.6 34.87 36.31 36.16 35.49 35.68 34.78 35.55 (db) 10&newJM 34.91 36.35 36.17 35.50 35.74 34.81 35.58 BitRate 5&JM8.6 121.25 37.65 45.27 65.25 124.06 34.13 71.27 (K@30HZ) 10&newJM 120.14 37.19 45.30 65.15 122.99 33.34 70.68 METime 5&JM8.6 68.17 44.20 35.36 41.78 59.39 37.97 47.81 (s) 10&newJM 6 3.16 39.08 28.56 35.85 52.12 31.43 41.70 | ||||||||
| SNRY 10&JM8.6 34.95 36.36 36.17 35.51 35.78 34.82 35.60 (db) 16&newJM 34.95 36.37 36.19 35.51 35.77 34.81 35.60 BitRate 10&JM8.6 119.75 37.10 45.27 65.03 122.64 33.23 70.50 (K@30HZ) 16&newJM 118.87 36.91 45.38 65.03 122.42 33.39 70.33 METime 10&JM8.6 146.07 89.94 67.22 83.71 139.10 70.85 99.48 (s) 16&newJM 75.18 45.69 33.44 42.58 60.55 37.59 49.17 | ||||||||
The cost that certain new algorithm is paid needs to increase memory space exactly, but for mobile device, the part increase that significantly reduces with respect to memory space of operand more is of practical significance.The another one characteristics of algorithm are, the complexity of algorithm can be because of the increase of reference frame linear increasing, after reference frame was increased to certain quantity, its ME can keep a more stable numerical value operation time.This can see from the realization principle of algorithm and the data of table three.
New algorithm can be seen by Fig. 4 and Fig. 3 equally to the improvement of RD performance.Fig. 3 be for the Carphone sequence adopt traditional JM algorithm of 5 reference frames respectively and adopt 10 reference frames new algorithm the RD curve ratio.Fig. 4 is for Mother﹠amp; The daughter sequence adopt traditional JM algorithm of 10 reference frames respectively and adopt 16 reference frames new algorithm the RD curve ratio.
In a word, the present invention proposes a new reference frame selection algorithm and be used for accelerating H.264 multiframe estimation.Based on the distribution of the RD expense of reference frame, and the inheritance of the RD expense between 7 kinds of predictive modes, we have used a reference frame preferable methods to come every kind of pattern is rebulid reference frame lists, thereby have saved a large amount of ME operation time.Analogue test confirmation new algorithm saved for 50% ME operation time, and the RD performance obtains the part improvement under the situation of using multi-reference frame (as 16 frames).Algorithm especially is fit to mobile video etc. in the demanding application of algorithm complex.The basis of method enforcement of the present invention and prior art is referring to as follows.
[1] " Draft ITU-T Recommendation and Fihal Draft International Standard ofJoint Video Specification (the H.264 draft standard that telecommunications alliance is recommended) (ITU-T Rec.H.264ISO/IEC 14496-10 AVC) ", Joint Video Team (JVT) of ISO/IEC MPEG and ITU-TVCEG, JVT-G050, Mar.2003.
Claims (2)
1, the adaptive reference frame selecting method in the multiframe estimation based on mode inheritance, it is characterized in that to lower level pattern estimation the time, using a new reference formation, the result of the construction basis higher level pattern estimation of this formation:, preferred reference frame is carried out estimation at 16 * 16,16 * 8,8 * 16,8 * 8,8 * 4,4 * 8 and 4 * 4 macro blocks of 7 kinds of inter prediction models and infra-frame prediction model;
The reference frame selection of following layer model is considered the inheritance that estimation predicts the outcome, and promptly the estimation based on upper mode predicts the outcome, and selects in the upper mode estimation more excellent several reference frames that predict the outcome to carry out estimation; At every kind of inheritance that pattern is concrete, before carrying out estimation, set up the reference frame preference lists of corresponding modes in the algorithm;
When carrying out the estimation of macro block, the order of model selection is to carry out according to descending order, i.e. 16 * 16 patterns at first, and 16 * 8 patterns, 8 * 16 patterns are 8 * 8 patterns and its subpattern at last then;
For 16 * 16 patterns, all candidate's reference frames are carried out estimation.For 16 * 8 patterns and 8 * 16 patterns, new reference frame lists comprises reference frame 1,2,3; Three reference frames that also comprise distortion rate RD expense minimum after 16 * 16 pattern estimation;
For 8 * 8 patterns, the reference frame lists of estimation has also comprised reference frame 4 and 5 with respect to 16 * 8 patterns and 8 * 16 patterns;
The motion estimation result of 16 * 8 and 8 * 16 patterns of correspondence when 8 * 4 and 4 * 8 schema reference frames are preferred; The reference frame lists of 4 * 4 patterns comprises three reference frames of the distortion rate RD expense minimum of reference frame 1,2,3 and 8 * 8 model predictions.
2, by the described adaptive reference frame selecting method of claim 1, it is characterized in that comprising reference frame 1,2,3 in the foundation of reference frame lists of 8 * 4 patterns and 4 * 8 patterns, three reference frames of the distortion rate RD expense minimum of 8 * 8 model predictions under also comprising also comprise three reference frames of the distortion rate RD expense minimum of 16 * 8 or 8 * 16 model predictions under it.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100943552A CN100362869C (en) | 2005-09-14 | 2005-09-14 | Adaptive reference frame selecting method based on mode inheritance in multiframe movement estimation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100943552A CN100362869C (en) | 2005-09-14 | 2005-09-14 | Adaptive reference frame selecting method based on mode inheritance in multiframe movement estimation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1741418A true CN1741418A (en) | 2006-03-01 |
| CN100362869C CN100362869C (en) | 2008-01-16 |
Family
ID=36093654
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2005100943552A Expired - Fee Related CN100362869C (en) | 2005-09-14 | 2005-09-14 | Adaptive reference frame selecting method based on mode inheritance in multiframe movement estimation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100362869C (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101170688B (en) * | 2007-11-26 | 2010-12-01 | 电子科技大学 | A quick selection method for macro block mode |
| CN101252692B (en) * | 2008-03-07 | 2011-05-18 | 炬力集成电路设计有限公司 | Apparatus and method for predicting between frames and video encoding and decoding eqiupment |
| CN102665079A (en) * | 2012-05-08 | 2012-09-12 | 北方工业大学 | Adaptive fast intra prediction mode decision for high efficiency video coding (HEVC) |
| CN104618715A (en) * | 2014-07-22 | 2015-05-13 | 腾讯科技（北京）有限公司 | Method and device for obtaining minimal rate-distortion cost |
| CN106303570A (en) * | 2016-08-22 | 2017-01-04 | 北京奇艺世纪科技有限公司 | A kind of Video coding reference frame selecting method and device |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2003281133A1 (en) * | 2002-07-15 | 2004-02-02 | Hitachi, Ltd. | Moving picture encoding method and decoding method |
| CN1323555C (en) * | 2005-01-26 | 2007-06-27 | 上海大学 | Quick selecting method for H.264/AVC multiple reference frame |
-
2005
- 2005-09-14 CN CNB2005100943552A patent/CN100362869C/en not_active Expired - Fee Related
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101170688B (en) * | 2007-11-26 | 2010-12-01 | 电子科技大学 | A quick selection method for macro block mode |
| CN101252692B (en) * | 2008-03-07 | 2011-05-18 | 炬力集成电路设计有限公司 | Apparatus and method for predicting between frames and video encoding and decoding eqiupment |
| CN102665079A (en) * | 2012-05-08 | 2012-09-12 | 北方工业大学 | Adaptive fast intra prediction mode decision for high efficiency video coding (HEVC) |
| CN102665079B (en) * | 2012-05-08 | 2014-11-26 | 北方工业大学 | Adaptive fast intra prediction mode decision for high efficiency video coding (HEVC) |
| CN104618715A (en) * | 2014-07-22 | 2015-05-13 | 腾讯科技（北京）有限公司 | Method and device for obtaining minimal rate-distortion cost |
| CN104618715B (en) * | 2014-07-22 | 2018-10-30 | 腾讯科技（北京）有限公司 | A kind of method and device obtaining minimum rate distortion costs |
| CN106303570A (en) * | 2016-08-22 | 2017-01-04 | 北京奇艺世纪科技有限公司 | A kind of Video coding reference frame selecting method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN100362869C (en) | 2008-01-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Huang et al. | Analysis and complexity reduction of multiple reference frames motion estimation in H. 264/AVC | |
| CN101015215B (en) | Methods and apparatus for performing fast mode decisions in video codecs. | |
| KR100359115B1 (en) | Video coding method | |
| JP4898467B2 (en) | Coding mode determination method and apparatus for variable block size motion prediction | |
| Kuo et al. | Fast variable block size motion estimation for H. 264 using likelihood and correlation of motion field | |
| US7764740B2 (en) | Fast block mode determining method for motion estimation, and apparatus thereof | |
| CN1719735A (en) | Method or device for coding a sequence of source pictures | |
| CN101888546B (en) | A kind of method of estimation and device | |
| KR20080067631A (en) | Mode selection techniques for multimedia coding | |
| CN105959690A (en) | Video encoding method for encoding division block, video decoding method for decoding division block, and recording medium for implementing the same | |
| CN101064849A (en) | Dynamic image coding method, apparatus and computer readable record medium | |
| CN1819658A (en) | Method and device for coding a video picture in inter or intra mode | |
| CN1879418A (en) | Selective and/or scalable complexity control for video codecs | |
| CN101331773B (en) | Device and method for processing rate controlled for video coding using rate-distortion characteristics | |
| CN100362869C (en) | Adaptive reference frame selecting method based on mode inheritance in multiframe movement estimation | |
| CN1325220A (en) | Motion-vector coding method | |
| CN1194544C (en) | Video encoding method based on prediction time and space domain conerent movement vectors | |
| CN1845608A (en) | Video interframe compression method based on space-time correlation | |
| Tian et al. | Adaptive intra mode decision for HEVC based on texture characteristics and multiple reference lines | |
| Wang et al. | Efficient coding tree unit (CTU) decision method for scalable high-efficiency video coding (SHVC) encoder | |
| KR101512643B1 (en) | 2 Video encoding apparatus and Apparatus and Method of 2 dimensional ordering transform for image signal and Recording Medium therefor | |
| CN106658024A (en) | Fast video coding method | |
| CN101977317B (en) | Intra-frame prediction method and device | |
| CN100396100C (en) | Motion estimation method of video compression | |
| CN1568009A (en) | Motion vector prediction method used for video coding |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C19 | Lapse of patent right due to non-payment of the annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |