CN100444639C - Method of selecting H.264/AVC frame to frame motion estimation mode - Google Patents
Method of selecting H.264/AVC frame to frame motion estimation mode Download PDFInfo
- Publication number
- CN100444639C CN100444639C CNB2005100867474A CN200510086747A CN100444639C CN 100444639 C CN100444639 C CN 100444639C CN B2005100867474 A CNB2005100867474 A CN B2005100867474A CN 200510086747 A CN200510086747 A CN 200510086747A CN 100444639 C CN100444639 C CN 100444639C
- Authority
- CN
- China
- Prior art keywords
- macro block
- mode
- max
- search
- select
- 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.)
- Expired - Fee Related
Links
Images
Landscapes
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The present invention relates to the video coding field and provides a mode selecting method for H. 264/AVC interframe motion estimation. According to the intense extent of the interframe motion of a macro block, the macro block is divided into modes of different sizes, an optimized prediction mode is obtained preliminarily and then according to the intense extent of the interframe motion of the macro block, optimal matching points are searched with different searching methods further to obtain the optimized prediction mode. Finally, the use rate distortion of the optimized prediction mode is optimized, and the optimal mode is determined. By using the method of the present invention, the operation quantity of encoding is largely reduced, and the searching time of the optimal matching point is largely shortened. The influence on image quality can be ignored, so an H. 264 standard obtains wider application.
Description
Technical field
The present invention relates to field of video encoding, be specifically related to the method for the selection of the selection of the interframe movement estimation model in the standard H.264/AVC and optimal match point.
Background technology
H.264 be by the common new video encoding standard of formulating of ISO/IEC and ITU-T two big standardization bodies.H.264/AVC the initial target of project is to wish that new codec can be under than the much lower code check of the video standard before relatively (such as MPEG-2 or H.263) (such as, half or still less) provides good video quality; Simultaneously, do not increase the coding tools of a lot of complexity, make that hardware is difficult to realize.The another one target is an adaptability, promptly this codec can in a very wide scope, use (such as, promptly comprise high code check and also comprise low code check, and different video resolution), and can be (such as multicast in diverse network and system, the DVD storage, RTP/IP packet network, ITU-T mediaphone system) work.
With H.263 comparing that people are familiar with, H.264 substandard video image quality is better, and code efficiency can improve more than 50% simultaneously, and has better fault-tolerant ability.This makes it be with a wide range of applications in the field of video communication in future.Yet owing to when image encoding, adopted multiframe and many macro blocks estimation (multi-frame and multi-block motionestimation), the complexity that makes coding calculate significantly increases, it is consuming time to increase coding, feasible network-adaptive variation H.264/AVC.
In order to improve code efficiency, in this standard, adopted many new technologies to improve coding efficiency.One of them most important technology is exactly rate-distortion optimization (RDO).In the prior art, when carrying out rate-distortion optimization, encoder will compare all possible macro block mode of video sequence interframe image, selects best pattern.This is a very big amount of calculation.Because, for one 16 * 16 macro block in the image, encoder can at first carry out 16 * 16 and 8 * 16 estimation (motion estimation) and rate-distortion optimization (RDO), this will need to calculate 2 rate distortion degree (rdcost), and 2 result calculated additions are obtained the Rdcost of one 16 * 16 macro block, select the preferred predictive mode of conduct of the Rdcost minimum of 16 * 16 macro blocks among the three again.And then 16 * 16 macro blocks are divided into 48 * 8 macro block, each 8 * 8 macro block is divided into 8 * 8,8 * 4,4 * 8,4 * 4 four kinds of patterns carries out estimation and RDO respectively.Wherein, 8 * 4 and 4 * 8 patterns obtain the Rdcost of 8 * 8 macro blocks with the Rdcost addition of twice calculating, and the Rdcost of 4 * 4 patterns also is added among the result.The Rdcost addition of 48 * 8 macro blocks obtains the Rdcost of one 16 * 16 macro block, selects the preferred predictive mode of conduct of 16 * 16 macro block Rdcost minimums from this four kind 8 * 8 is divided.Compare these 2 preferred predictive modes that obtained at last, the inter-frame forecast mode of conduct the best that the selection rate distortion value is less.Encoder need repeat above step to each macro block of a video sequence, is all determined up to the best inter mode of all macro blocks of sequence.
So as can be seen coding techniques, operand are very huge.In the prior art 16 * 16 macro blocks being carried out the execution number of times that estimation and Rdcost calculate is: 16 * 16 once, 16 * 8 twice and 8 * 16 twice.The execution number of times that the estimation of 8 * 8 macro blocks and rate distortion value are calculated is 8 * 81 * 4=4 time, 8 * 42 * 4=8 time, and 4 * 82 * 4=8 time, 4 * 44 * 4=16 time.The execution number of times that one 16 * 16 macro block needs the estimation of carrying out and Rdcost to calculate altogether is: 1+2+2+4+8+8+16=41 time.For the image of QCIF form, a frame has 99 16 * 16 macroblocks, so the execution number of times that the two field picture of encoding will be finished estimation and Rdcost calculating is 41 * 99=4059 time.Wherein, the big height of the calculation of complex of Rdcost.Repeat the high calculating of such complexity in a large number, its coding is very consuming time, and is very big to the consumption of resource, also is not easy to real-time application.
In the prior art, using this conventional implementation method of calculating is JM software.JM be the expert group of ITU-T in formulating the process of standard, follow standard that corresponding test model is provided aperiodically.These test models adopt standard C language to write usually, and main purpose is to test for the correctness of verifying encoder and to coding efficiency.Therefore, although rate-distortion optimization can bring extraordinary coding efficiency, the search of all patterns consumes the plenty of time during for coding, and this makes that H.264 encoder is difficult to real time business.The method of a lot of estimation fast and pattern decision is suggested and is used to improve this situation, in " quick mode among the JVT/H.264 is selected and estimation " of delivering in " the quick realization of many macro blocks estimation in a kind of new H.264 video coding " delivered in multimedia international conference (ICME2003) in 2003 and the image processing international conference in 2003, propose a kind of algorithm that utilizes the threshold value premature termination, be used for reducing search for candidate's predictive mode.
These methods that realize fast are defectiveness still.Be mainly reflected in for the selection of preferred predictive mode still too complicatedly, need carry out the search and relative more calculating of sub-pix point.
In addition, the search method that needs the optimal match point used in the cataloged procedure, for example now more typical three step search methods (TSS), diamond search method (DS), hexagon search method (HEXBS) etc., in the time of on specifically being applied to H.264 whole pixel searching algorithm, still defectiveness, main performance be in the following areas:
1, in standard H.264, in order to improve the estimation effect, adopted a plurality of reference frames to participate in the algorithm of estimating, for same macro block, adopt seven kinds of different partitioned modes.In the motion estimation process of a macro block, must be to each reference frame, each divides block mode and each to divide each piece under the block mode to carry out motion estimation search respectively, to calculate the result of optimal match point as estimation.Such coding computational process is complicated and consuming time.
Search radius is fixed when 2, searching for optimal match point.Because for evenly fixing image-region, excessive search radius is little for improving the effect of search performance, can not reach optimal effectiveness when therefore searching for, and resource is caused certain waste.
3, diamond search method and hexagon search method, for being shaped as foursquare block mode, coverage effect is relatively good, (the experiment statistics data show because its hunting zone is similar to a circle, it is being in 2 the circle with search radius that 50%～90% motion vector is concentrated), but for 16 * 8,8 * 16,8 * 4,4 * 8 block modes, because the length and width of piece are inconsistent, its search coverage effect is bad.
Summary of the invention
(1) technical problem that will solve
The objective of the invention is to overcome the defective of present prior art, propose a kind of efficient higher, be applied to method for video coding H.264/AVC standard, that can realize fast, the inter-frame mode that solves estimation when encoding in the prior art is selected too high, the consuming time too much problem of computation complexity, makes that H.264/AVC the video coding of standard can have more real-time application.
(2) technical scheme
The present invention proposes to determine in a kind of H.264/AVC interframe movement estimation the method for optimal match point, may further comprise the steps:
(4) video sequence is divided into a plurality of macro blocks;
(5) the absolute difference sum of computing macro block interframe corresponding pixel points is SAD, according to the interframe movement severe degree of SAD judgment frame macro block;
(6) macro block different to the motion severe degree is divided into large scale macro block, middle size macro block, small size macro block, and the inviolent macroblock partitions of moving is the large scale macro block, and size is 16 * 16,16 * 8,8 * 16; Size macro block during the relatively more significant macro block of motion is divided into again, size is 8 * 8,8 * 4,4 * 8; The macroblock partitions of motion highly significant is the small size macro block, and size is 4 * 4;
(4) at the macro block of the different size of step (3),, use diamond search pattern to determine the images match pixel for the large scale macro block; For middle size macro block, select for use the hexagon search pattern to determine preliminary match point earlier, be the center with this point again, use diamond search pattern to carry out motion search, determine optimal match point; For the small size macro block, use big hexagon search pattern to search for earlier, find the scope of motion vector indication, select for use the hexagon search pattern to search for again, find preliminary match point, putting with this is search center again, uses diamond search pattern to determine the matched pixel point.
The present invention also proposes a kind of method of model selection of H.264/AVC interframe movement estimation, and it may further comprise the steps:
(1) two field picture of video sequence is divided into a plurality of macro blocks;
(2) calculate the SAD of corresponding picture element between each macroblock frame;
(3) set threshold T H
1, with the sad value of step (2) gained and the threshold T H that sets
1Compare, promptly for sad value less than TH
1Situation, use big macro block mode, for sad value greater than TH
1Situation, the middle or small macro block mode of use;
(4), use diamond search pattern to determine the images match pixel to described big macro block mode; For middle macro block mode, select for use the hexagon search pattern to determine preliminary match point earlier, be the center with this point again, use diamond search pattern to carry out motion search, determine optimal match point; For little macro block mode, use big hexagon search pattern to search for earlier, find the scope of motion vector indication, select for use the hexagon search pattern to search for again, find preliminary match point, putting with this is search center again, uses diamond search pattern to determine the matched pixel point;
(5) the adjacent and vertical adjacent sad value of totally 4 some pixels of level of the optimal match point determined of calculation procedure (4), wherein the sad value of horizontal pixel point is labeled as SADH, and the sad value of vertical pixel point is labeled as SADV;
(6) 4 sad values to gained in the step (5) compare, and select preferred predictive mode according to different comparative results;
(7) the preferred predictive mode that step (6) is obtained carries out rate-distortion optimization RDO, determines optimal mode.
Wherein, the big macro block mode in the step (3) is 16 * 16,16 * 8,8 * 16 patterns; Middle macro block mode is 8 * 8,8 * 4,4 * 8 patterns; Little macro block mode is 4 * 4 patterns.
Wherein, for step (6), can be labeled as SADH to bigger value among 2 SADH
Max, bigger value among 2 SADV is labeled as SADV
MaxUnder the situation of big macro block mode, for SADH
Max＞SADV
MaxSituation, select 16 * 8 and 16 * 16 patterns as preferred predictive mode; For SADH
Max＜SADV
MaxSituation, select 8 * 16 and 16 * 16 patterns as preferred predictive mode; For SADH
Max=SADV
MaxSituation, select for use 16 * 8,8 * 16 and 16 * 16 patterns as preferred predictive mode simultaneously.
Wherein, for step (6), under the situation of middle or small macro block mode, situation for SADHmax＞SADVmax, 8 * 4 and 8 * 8 patterns of selection for the situation of SADHmax＜SADVmax, select 4 * 8 and 8 * 8 patterns as preferred predictive mode as preferred predictive mode; For the situation of SADHmax=SADVmax, select for use 8 * 4 and 4 * 8 and 8 * 8 patterns simultaneously as preferred predictive mode.
(3) beneficial effect
Use the method for the quick realization of model selection that H.264/AVC interframe movement of the present invention estimates and estimation, can obtain following advantage:
1, because main computing of the present invention is to calculate SAD, and SAD calculates easily, so reduced the amount of calculation in the cataloged procedure greatly, economizes on resources.
2, owing to there is very most even and steady part in the image sequence, the chance that these parts are divided into the small size macro block is very little, and there is no need to carry out the RDO calculating of small size macro block (mb) type again.Therefore, different image sections is adopted different macroblock size, can save the scramble time greatly and reduce encoder complexity, make coding method adapt to the requirement of real time business.
3, experimental result shows, using the JM9.3 method to calculate in method of the present invention and the prior art compares, for different quantization steps, estimation that can decreased average 70.4% is consuming time, and aspect the picture quality influence, picture quality is the minimizing of PSNR only about 0.1dB, and such loss is negligible, and the increase of output code flow bit rate is only in 2.5%.That is to say, adopt method of the present invention after, when having improved code efficiency greatly, but be very little to the negative effect of picture quality.
Description of drawings
Fig. 1 is in the method for the model selection estimated of interframe movement of the present invention, selects the flow chart of preferred predictive mode;
Fig. 2 is in the method for the model selection estimated of interframe movement of the present invention, determines the flow chart of optimal mode in selected preferred predictive mode;
Fig. 3 is the schematic diagram of an optimal match point and adjacent 4 pixels thereof;
Fig. 4 is among the embodiment, adopts the method for method of the present invention and prior art, in the comparison of the number of the pixel of searching for when determining optimal match point;
Fig. 5 is the effect contrast figure of embodiment 3;
Fig. 6 is the effect contrast figure of embodiment 4;
Fig. 7 is the effect contrast figure of embodiment 5.
Embodiment
The method of the model selection that the H.264/AVC interframe movement that the present invention proposes is estimated is described as follows in conjunction with the accompanying drawings and embodiments.Following execution mode only is used to illustrate the present invention; and be not limitation of the present invention; the those of ordinary skill in relevant technologies field; under the situation that does not break away from the spirit and scope of the present invention; can also make various variations and modification; therefore all technical schemes that are equal to also belong to category of the present invention, and scope of patent protection of the present invention should be limited by each claim.
Below the experiment condition of each embodiment be:
(1) preceding 100 frames of test video sequence is encoded;
(2) adopt CAVLC (based on contextual variable-length encoding) entropy coding;
(3) adopt 5 reference frames;
(4) hunting zone is 32 pixels;
(5) adopt Hadamard (Hadamard) conversion;
(6) utilization rate aberration optimizing (RDO);
(7) coded sequence GOP is IPPP;
(8) experiment use JM9.3 software for contrast with reference to (be JVT Reference Softwareunofficial version JM9.3, the software download address is http://bs.hhi.de/～suehring/tml/download);
(9) computer configuration of experiment employing is the CPU of P4Celeron 2.66GHz, the internal memory of 256M.
In following examples the standard video sequence of various QCIF, CIF form is tested, the result of contrast dual mode: wherein, a kind of is to adopt original setting of JM9.3 promptly to adopt the method for search fast that whole macro block modes are encoded, and another kind is that the method for the model selection estimated of interframe movement that this paper proposes is encoded.
(embodiment one)
In the present embodiment, to 5 video sequences, be respectively " container ", " news ", " foreman ", " tempete " and " pairs " tests, the contrast prior art and adopt method of the present invention after, the change of the calculation times of estimation and RDO.
Because the prior art temporal correlation between the considered frame image not, so for the video sequence of QCIF form, every frame has 99 16 * 16 macro block, therefore every frame all needs to carry out 4056 estimation and the rate distortion value is calculated.
After adopting method mode system of selection of the present invention, the operation times of each video sequence is as shown in table 1:
| The sequence title | The calculation times of estimation and rate-distortion optimization |
| container | 552 |
| news | 796 |
| foreman | 661 |
| tempete | 1491 |
| pairs | 1265 |
Table 1
As seen, because in the prior art, when video sequence was carried out rate-distortion optimization, encoder will compare all possible macro block mode of inter frame image, select best pattern, so operand is very big; And adopt method of the present invention, and can judge earlier the motion severe degree of inter frame image tentatively to obtain preferred pattern, reduce the number of times of operation RDO for the first time; Use different searching methods to determine optimal match point according to the violent program of the motion of inter frame image then, the sad value that the calculating optimum match point is adjacent has further reduced preferred number of modes according to these values, at last to these preferred mode operation RDO.Therefore, through such 2 minimizings to the predictive mode number, for the not too violent video sequence of motion as " container ", operation times significantly reduces to original about 1/7, even and the worst situation, the very violent video sequence " tempete " of motion just, its operation times also is reduced to 1491 times, and very large amount of decrease is arranged compared to existing technology.
Fig. 4 has shown the method that adopts method of the present invention and prior art, in the comparison of the number of the pixel of searching for when determining optimal match point.The post of wherein ruling is represented the pixel search number of the method for prior art, do not have the line post and represent to adopt the pixel search number of method of the present invention, transverse axis is represented 5 test video sequence that present embodiment uses, and the longitudinal axis is represented the comparison of the number of the pixel searched for when determining optimal match point.As seen adopt method of the present invention that cycle tests is encoded,, significantly reducing aspect the search pixel point number,, can reduce the volumes of searches of half reasonable situation compared with prior art.
(embodiment two)
In the present embodiment, at one 16 * 16 test macro block, the JM9.3 of contrast prior art and the Fast implementation of estimation of the present invention are under the pattern of 7 kinds of different sizes, the number and the number of times of the pixel of required search during contrast calculating optimum match point, and the amount of calculation.
In the prior art, motion conditions is not classified in the search of match point, pattern for 16 * 16 at first adopts the hexagon search pattern, adopt diamond search pattern then, and all at first use big hexagon pattern for all the other all patterns, be to adopt hexagon pattern and diamond pattern searching method more then, so amount of calculation is huge.
In the present embodiment, consider the movement relation of interframe, different macro blocks is adopted different searching methods, therefore, need search for 152 pixels for big hexagon pattern, the pixel number that the hexagon pattern need be searched for is 192, and diamond pattern then is 128.
Table 2 provided adopt method of the present invention and prior art JM 9.3 methods for one 16 * 16 macro block, the contrast of the number of the pixel that search for and number of times.The JM9.3 method will be searched for 320 points for 16 * 16 as can be seen, and as long as method of the present invention is 128; For all the other six kinds of patterns, the prior art search point all is 472, and method of the present invention can be selected different macroblock partitions according to the image motion situation, therefore divides the desired difference of counting for every kind, and divides more that search point reduces for large scale.
Table 2
(embodiment three)
In the present embodiment, respectively to 9 video sequences, standard video sequence comprising QCIF form and CIF form is tested, the result of contrast dual mode: a kind of is to adopt the original setting of JM9.3 promptly to adopt rapid motion estimating method that whole macro block modes are encoded, and another kind is the method for the quick realization of the model selection estimated of interframe movement that the present invention proposes and estimation.
In order to obtain effect preferably, guarantee coding efficiency, for threshold T H
1Selection extremely important.Select according to being to find a balance preferably between consuming time and output code flow bit rate, the picture quality PSNR in motion compensation.
Learn different TH by repeatedly testing
1The minimizing consuming time of estimation under the condition, picture quality changes all different with stream rate, takes all factors into consideration to allow picture quality reduce and output code flow bit rate variation minimum.For example in the present embodiment, be under 28 the situation, to select 6000 at quantization parameter QP as thresholding.In application backward, if the user can change this threshold value, to obtain suitable effect to the picture quality or the specific (special) requirements that has consuming time of encoding.
Table 3 has provided under quantization parameter QP=28 situation, and the test result to 9 different sequences comprises the contrast prior art, adopt method of the present invention after, the variation of estimation time, the wherein saving of negative sign express time; The variation of picture quality, wherein negative sign is represented distortion; Variation with the output code flow bit rate.
| Sequence title (form) | The estimation time changes (%) | Picture quality changes (dB) | Bit rate changes (%) |
| news(QCIF) | -72.992 | -0.060 | 2.436 |
| salesman(QCIF) | -74.448 | -0.200 | 1.890 |
| carphone(QCIF) | -67.226 | -0.100 | 3.857 |
| foreman(QCIF) | -55.938 | -0.060 | 6.081 |
| claire(QCIF) | -57.290 | -0.140 | 1.256 |
| container(QCIF) | -69.313 | -0.060 | 0.037 |
| mother&daughter(QCIF) | -84.840 | -0.110 | 1.563 |
| mobile(CIF) | -46.944 | -0.040 | 2.836 |
| tempete(CIF) | -48.234 | -0.020 | 2.763 |
| Average | -64.136 | -0.088 | 2.524 |
Table 3
Fig. 5 has shown that the method that adopts method of the present invention and prior art is in estimation comparison consuming time.The post of wherein ruling represents that the method estimation of prior art is consuming time, do not have the line post and represents to adopt the consuming time of method of the present invention, and 9 standard video sequence above 1～9 order expression of transverse axis numeral in the table, the longitudinal axis is represented the required time of estimation, unit is second.
As seen, for QP=28, TH
1=6000 situation, the image sequence estimation time can be saved 84.8% (" mother and daughter " sequence) at most stably, also can improve 46.9% (" mobile " sequence) for the violent image sequence minimum of motion, it is consuming time on average can to save 64.1% estimation
(embodiment four)
In the present embodiment, respectively to 9 video sequences, standard video sequence comprising QCIF form and CIF form is tested, the result of contrast dual mode: a kind of is to adopt the original setting of JM9.3 promptly to adopt rapid motion estimating method that whole macro block modes are encoded, and another kind is the method for the quick realization of the model selection estimated of interframe movement that the present invention proposes and estimation.
Quantization parameter QP in the present embodiment is 32, and the threshold value of selection is 10000.
Table 4 has provided the test result to 9 different sequences, comprises the contrast prior art, adopt method of the present invention after, the variation of estimation time, the wherein saving of negative sign express time; The variation of picture quality, wherein negative sign is represented distortion; Variation with the output code flow bit rate.
| Sequence title (form) | The estimation time changes (%) | Picture quality changes (dB) | Bit rate changes (%) |
| news(QCIF) | -86.679 | -0.090 | 1.782 |
| salesman(QCIF) | -82.447 | -0.080 | 2.293 |
| carphone(QCIF) | -75.258 | -0.120 | 2.083 |
| foreman(QCIF) | -64.838 | -0.200 | 3.548 |
| claire(QCIF) | -65.134 | 0.190 | 3.240 |
| container(QCIF) | -69.750 | -0.070 | 0.249 |
| mother&daughter(QCIF) | -81.884 | -0.010 | 2.261 |
| mobile(CIF) | -61.214 | -0.070 | 0.350 |
| tempete(CIF) | -52.589 | -0.060 | 1.162 |
| Average | -71.088 | -0.057 | 1.885 |
Table 4
As seen, for QP=32, TH
1=10000 situation, the estimation time maximum of image sequence can be saved 86.7% (" news " sequence) stably, also can improve 52.6% (" tempete " sequence) for the violent image sequence minimum of motion, it is consuming time on average can to save 71% estimation, and the loss of picture quality on average has only 0.06 decibel, can ignore.
Fig. 6 has shown that the method that adopts method of the present invention and prior art is in estimation comparison consuming time.The post of wherein ruling represents that the method estimation of prior art is consuming time, do not have the line post and represents to adopt the consuming time of method of the present invention, and 9 standard video sequence above 1～9 order expression of transverse axis numeral in the table, the longitudinal axis is represented the required time of estimation, unit is second.
(embodiment five)
In the present embodiment, respectively to 9 video sequences, standard video sequence comprising QCIF form and CIF form is tested, the result of contrast dual mode: a kind of is to adopt the original setting of JM9.3 promptly to adopt rapid motion estimating method that whole macro block modes are encoded, and another kind is the method for the quick realization of the model selection estimated of interframe movement that the present invention proposes and estimation.
Quantization parameter QP in the present embodiment is 36, and the threshold value of selection is 14000.
Table 5 has provided the test result to 9 different sequences, comprises the contrast prior art, adopt method of the present invention after, the variation of estimation time, the wherein saving of negative sign express time; The variation of picture quality, wherein negative sign is represented distortion; Variation with the output code flow bit rate.
| Sequence title (form) | The estimation time changes (%) | Picture quality changes (dB) | Bit rate changes (%) |
| news(QCIF) | -82.612 | -0.150 | 2.104 |
| salesman(QCIF) | -84.703 | -0.130 | 0.850 |
| carphone(QCIF) | -81.861 | -0.110 | 1.196 |
| foreman(QCIF) | -77.149 | -0.280 | 5.219 |
| claire(QCIF) | -61.725 | -0.210 | 4.267 |
| container(QCIF) | -73.491 | -0.020 | 0.808 |
| mother&daughter(QCIF) | -83.910 | 0.020 | 2.544 |
| mobile(CIF) | -63.528 | -0.060 | 0.237 |
| tempete(CIF) | -74.287 | 0.000 | 0.029 |
| Average | -75.919 | -0.104 | 1.917 |
Table 5
As seen, for QP=36, TH
1=14000 situation, the estimation time maximum of image sequence can be saved 84.7% (" salesman " sequence) stably, also can improve 61.7% (" claire " sequence) for the violent image sequence minimum of motion, it is consuming time on average can to save 76% estimation, and the loss of picture quality on average has only 0.1 decibel, can ignore.
Fig. 7 has shown that the method that adopts method of the present invention and prior art is in estimation comparison consuming time.The post of wherein ruling represents that the method estimation of prior art is consuming time, do not have the line post and represents to adopt the consuming time of method of the present invention, and 9 standard video sequence above 1～9 order expression of transverse axis numeral in the table, the longitudinal axis is represented the required time of estimation, unit is second.
Claims (5)
1, determine the method for optimal match point during a kind of H.264/AVC interframe movement is estimated, it is characterized in that, may further comprise the steps:
(1) video sequence is divided into a plurality of macro blocks;
(2) the absolute difference sum of computing macro block interframe corresponding pixel points is SAD, according to the interframe movement severe degree of SAD judgment frame macro block;
(3) macro block different to the motion severe degree is divided into large scale macro block, middle size macro block, small size macro block, and the inviolent macroblock partitions of moving is the large scale macro block, and size is 16 * 16,16 * 8,8 * 16; Size macro block during the relatively more significant macro block of motion is divided into again, size is 8 * 8,8 * 4,4 * 8; The macroblock partitions of motion highly significant is the small size macro block, and size is 4 * 4;
(4) at the macro block of the different size of step (3),, use diamond search pattern to determine the images match pixel for the large scale macro block; For middle size macro block, select for use the hexagon search pattern to determine preliminary match point earlier, be the center with this point again, use diamond search pattern to carry out motion search, determine optimal match point; For the small size macro block, use big hexagon search pattern to search for earlier, find the scope of motion vector indication, select for use the hexagon search pattern to search for again, find preliminary match point, putting with this is search center again, uses diamond search pattern to determine the matched pixel point.
2, a kind of method of model selection of H.264/AVC interframe movement estimation is characterized in that it may further comprise the steps:
(1) two field picture of video sequence is divided into a plurality of macro blocks;
(2) calculate the SAD of corresponding picture element between each macroblock frame;
(3) set threshold T H
1, with the sad value of step (2) gained and the threshold T H that sets
1Compare, promptly for sad value less than TH
1Situation, use big macro block mode, for sad value greater than TH
1Situation, the middle or small macro block mode of use;
(4), use diamond search pattern to determine the images match pixel to described big macro block mode; For middle macro block mode, select for use the hexagon search pattern to determine preliminary match point earlier, be the center with this point again, use diamond search pattern to carry out motion search, determine optimal match point; For little macro block mode, use big hexagon search pattern to search for earlier, find the scope of motion vector indication, select for use the hexagon search pattern to search for again, find preliminary match point, putting with this is search center again, uses diamond search pattern to determine the matched pixel point;
(5) the adjacent and vertical adjacent sad value of totally 4 some pixels of level of the optimal match point determined of calculation procedure (4), wherein the sad value of horizontal pixel point is labeled as SADH, and the sad value of vertical pixel point is labeled as SADV;
(6) 4 sad values to gained in the step (5) compare, and select preferred predictive mode according to different comparative results;
(7) the preferred predictive mode that step (6) is obtained carries out rate-distortion optimization RDO, determines optimal mode.
3, method as claimed in claim 2 is characterized in that, the big macro block mode in the step (3) is 16 * 16,16 * 8,8 * 16 patterns; Middle macro block mode is 8 * 8,8 * 4,4 * 8 patterns; Little macro block mode is 4 * 4 patterns.
4, method as claimed in claim 3 is characterized in that, for step (6), can be labeled as SADH to bigger value among 2 SADH
Max, bigger value among 2 SADV is labeled as SADV
MaxUnder the situation of big macro block mode, for SADH
Max＞SADV
MaxSituation, select 16 * 8 and 16 * 16 patterns as preferred predictive mode; For SADH
Max＜SADV
MaxSituation, select 8 * 16 and 16 * 16 patterns as preferred predictive mode; For SADH
Max=SADV
MaxSituation, select for use 16 * 8,8 * 16 and 16 * 16 patterns as preferred predictive mode simultaneously.
5, method as claimed in claim 4 is characterized in that, for step (6), under the situation of middle or small macro block mode, for SADH
Max＞SADV
MaxSituation, select 8 * 4 and 8 * 8 patterns as preferred predictive mode, for SADH
Max＜SADV
MaxSituation, select 4 * 8 and 8 * 8 patterns as preferred predictive mode; For SADH
Max=SADV
MaxSituation, select for use 8 * 4 and 4 * 8 and 8 * 8 patterns as preferred predictive mode simultaneously.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100867474A CN100444639C (en) | 2005-10-31 | 2005-10-31 | Method of selecting H.264/AVC frame to frame motion estimation mode |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100867474A CN100444639C (en) | 2005-10-31 | 2005-10-31 | Method of selecting H.264/AVC frame to frame motion estimation mode |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1753501A CN1753501A (en) | 2006-03-29 |
| CN100444639C true CN100444639C (en) | 2008-12-17 |
Family
ID=36680150
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2005100867474A Expired - Fee Related CN100444639C (en) | 2005-10-31 | 2005-10-31 | Method of selecting H.264/AVC frame to frame motion estimation mode |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100444639C (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100471275C (en) * | 2006-09-08 | 2009-03-18 | 清华大学 | Motion estimating method for H.264/AVC coder |
| CN101022555B (en) * | 2007-02-12 | 2010-12-08 | 清华大学 | Interframe predictive coding mode quick selecting method |
| CN100558169C (en) * | 2007-03-23 | 2009-11-04 | 武汉大学 | The fast selecting method of inter-frame mode in the video code conversion |
| US8467451B2 (en) * | 2007-11-07 | 2013-06-18 | Industrial Technology Research Institute | Methods for selecting a prediction mode |
| CN101267556B (en) * | 2008-03-21 | 2011-06-22 | 海信集团有限公司 | Quick motion estimation method and video coding and decoding method |
| CN101309424B (en) * | 2008-07-04 | 2011-11-30 | 四川虹微技术有限公司 | Quick movement estimating method |
| CN101605262B (en) * | 2009-07-09 | 2011-11-09 | 杭州士兰微电子股份有限公司 | Method and device for predicting size motion of variable block |
| CN102843556B (en) * | 2011-06-20 | 2015-04-15 | 富士通株式会社 | Video coding method and video coding system |
| CN103079068A (en) * | 2013-01-14 | 2013-05-01 | 天津大学 | Interframe mode judging method based on relevance between motion intensity and space domain |
| CN109547798B (en) * | 2018-12-17 | 2020-09-18 | 杭州当虹科技股份有限公司 | Rapid HEVC inter-frame mode selection method |
| CN112492314A (en) * | 2020-11-25 | 2021-03-12 | 杭州微帧信息科技有限公司 | Method for selecting dynamic motion estimation algorithm based on machine learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6175593B1 (en) * | 1997-07-30 | 2001-01-16 | Lg Electronics Inc. | Method for estimating motion vector in moving picture |
| CN1556652A (en) * | 2003-12-30 | 2004-12-22 | 中国科学院计算技术研究所 | Video freguencying frame image fast coding method based on optimal prediction mode probability |
| CN1694535A (en) * | 2005-06-20 | 2005-11-09 | 浙江大学 | Complexity hierarchical mode selection method |
-
2005
- 2005-10-31 CN CNB2005100867474A patent/CN100444639C/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6175593B1 (en) * | 1997-07-30 | 2001-01-16 | Lg Electronics Inc. | Method for estimating motion vector in moving picture |
| CN1556652A (en) * | 2003-12-30 | 2004-12-22 | 中国科学院计算技术研究所 | Video freguencying frame image fast coding method based on optimal prediction mode probability |
| CN1694535A (en) * | 2005-06-20 | 2005-11-09 | 浙江大学 | Complexity hierarchical mode selection method |
Non-Patent Citations (4)
| Title |
|---|
| H.264帧间编码模式选择快速算法. 何莉莉.西华大学学报自然科学版，第24卷第4期. 2005 |
| H.264帧间编码模式选择快速算法. 何莉莉.西华大学学报自然科学版，第24卷第4期. 2005 * |
| H.264快速帧间编码模式选择算法. 朱冬冬，戴琼海.有线电视技术，第2004年第9期（总第153期）期. 2004 |
| H.264快速帧间编码模式选择算法. 朱冬冬，戴琼海.有线电视技术，第2004年第9期（总第153期）期. 2004 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1753501A (en) | 2006-03-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100444639C (en) | Method of selecting H.264/AVC frame to frame motion estimation mode | |
| CN100571390C (en) | A kind of H264 video coding fast schema selection method and device | |
| CN101640802B (en) | Video inter-frame compression coding method based on macroblock features and statistical properties | |
| CN100417228C (en) | Method of selecting in frame prediction mode based on H.264/AVC standard frame image | |
| CN102025995B (en) | Spatial enhancement layer rapid mode selection method of scalable video coding | |
| CN101394566B (en) | Cross rhombic motion estimation searching method | |
| CN100574447C (en) | Fast intraframe predicting mode selecting method based on the AVS video coding | |
| CN103517069A (en) | HEVC intra-frame prediction quick mode selection method based on texture analysis | |
| CN103997646A (en) | Rapid intra-frame prediction mode selection method in high-definition video coding | |
| CN103384325A (en) | Quick inter-frame prediction mode selection method for AVS-M video coding | |
| CN101888546B (en) | A kind of method of estimation and device | |
| CN101304529A (en) | Method and device for selecting macro block pattern | |
| CN103384326A (en) | Quick intra-frame prediction mode selection method for AVS-M video coding | |
| CN105657420A (en) | HEVC-oriented fast intra-frame prediction mode decision method and device | |
| CN103384327A (en) | AVS fast mode selection algorithm based on adaptive threshold | |
| CN104333754A (en) | SHVC (scalable high efficiency video coding) enhancement layer video coding method based on rapid prediction mode selection | |
| CN101754022A (en) | Motion estimation method with low complexity | |
| CN101179729A (en) | Interframe mode statistical classification based H.264 macroblock mode selecting method | |
| CN101237580A (en) | Integer pixel quick mixing search method based on center prediction | |
| CN101883275B (en) | Video coding method | |
| CN101426140B (en) | AVS fast schema selection method and device | |
| CN1194544C (en) | Video encoding method based on prediction time and space domain conerent movement vectors | |
| CN102158702A (en) | Self-adaption H.264 code rate control method | |
| CN101867818B (en) | Selection method and device of macroblock mode | |
| WO2007089916A2 (en) | Dynamic reference frame decision method and system |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20081217 Termination date: 20181031 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |