H.264-based Motion Estimation And Inter-mode Selection Algorithm

Since CCITT proposed the draft standard of H.261in1980s, driven by computer technologyand multimedia technology, video coding technology has gained rapid development. H.264/AVCstandard draft was put forward in2003, inherited the codec structure of the past video codingstandard, but introduced a number of advanced technologies. Multiple reference frames, and avariety of block partition inter prediction techniques greatly improve the accuracy of prediction,but bring additional coding load. Inter-frame redundancy is the main redundancy in videosequences, inter-frame prediction encoding also occupies most of the encoding time of the videocoding. Therefore, inter-frame coding technology has a high research value.Motion estimation and inter-mode selection are two main parts in inter-frame predictioncoding technology. The major computation of inter-frame prediction encoding comes frommotion estimation, the traditional motion estimation algorithms are hard to get an accurate motionvector with less search points. Traverse inter mode selection needs to search motion vectors forevery inter mode. H.264/AVC provides7kinds of inter prediction modes for inter frame macroblock, thus doubles the computation of motion estimation. Therefore, this thesis chose motionestimation and mode selection, the two contents of H.264/AVC to do the thorough research, andproposes improved algorithms for each content.Although UMHexagonS algorithm has high search efficiency, there are still two defects inthe algorithm. Firstly, mixed search templates exists repeated search points. Secondly, unevenmulti-hexagon-grid search template exits amount of search redundancy for small motion blocks.Aiming at the repeating search problem, we designed an enhanced big diamond template toreplace the5×5template in the center, and by using the complementarity and the search rulesanalysis between search patterns so as to avoid the repeating search between templates; Aiming toreduce the search redundancy for small motion blocks, we designed an early termination strategyin uneven multi-hexagon-grid search progress, which is only related to search process, butspacial-temple correlation. The experimental results indicate that, the improved algorithm and theoriginal algorithm have the same reconstruction image quality, with only negligible change ofPSNR(Y) and output stream bit rate, the improved algorithm can effectively reduces searchpoints and motion estimation time. The improved algorithm effectively enhances codingefficiency, especially for complex texture and high motion video sequences.Based on the analysis of8×8blocks’ motion vector homogeneity, we designed a fastinter-frame macro block mode selection algorithm. The algorithm selects the macro block modedirectly by using8×8blocks’ motion vector homogeneity judgment factors. In order to get theaccurate motion vector of8×8blocks, we deeply analysed the modes correlation in motion estimation progress, such as block matching correlation, search windows correlation and thecoarse positioning template search points correction in UMHexagonS. According to thesecorrections, we designed an8×8block’s motion vector positioning algorithm which positioning8×8blocks’ motion vectors in the progress of16×16mode’s motion vector search. Ourpositioning algorithm can not only accurately positioning the regions of8×8blocks’ optimalmotion vector, but also save the search time of coarse positioning template in original algorithmfor8×8blocks. Additionally we proposed the concept of down layer prediction. Aiming at theup layer modes of8×8mode (mode2and mode3), if one of the mode’s motion vectorhomogeneity judgment factor is minimum, we select this mode as the optimal mode, and predictthe selected mode’s motion vector by using down layer prediction motion vector, so as to skip theselected mode’s coarse positioning template search points. The experimental results demonstratethe accuracy of our fast inter mode selection algorithm based on8×8blocks’ motion vectorhomogeneity analysis, the accuracy of motion vector positioning algorithm based on modescorrelation, and the high efficiency of our fast mode selection algorithm based on UMHexagonS.