Motion Estimation In Video Compression Algorithm Optimization And Research

Motion estimation (ME) plays an important role in video compressionalgorithms. It is the key technique to improve coding efficiency because of itsstrong competence to reduce the temporal redundancy. The traditional full search(FS) motion estimation algorithm needs to calculate the matching error of everycandidate point within search window, so its computation complexity is verylarge. The latest video coding standard H.264/AVC introduces several newtechniques to improve the performance of motion estimation module, namely,variable block size matching, quarter pixel accuracy of motion vector andmultiple reference frame prediction. These new techniques enhance the matchingaccuracy of the searching blocks, on the other hand, the computation complexityof motion estimation module dramatically increases.Aiming at reducing the computation complexity of the motion estimationmodule, the work and research of this paper are done and listed as follow. Inorder to reduce the complexity, a lot of excellent algorithms have been proposedin recent years. They are comprehensively reviewed and researched in this paper,and a systemic classification is presented according to core ideas of thesealgorithms. After that, two novel motion estimation algorithms are proposed onthe basis of design principle of motion estimation module. Firstly, aiming atoptimizing the motion predictor selection process of EPZS algorithm, the sortingprocess for these predictors has been proposed. In the predictor sorting basedalgorithm, the two other key techniques are composed together, namely, dynamicsearch range and adaptively changed search patterns, as well as the earlytermination (ET) technique derived from EPZS. The second novel algorithmproposed in this paper is Meanshift based motion estimation algorithm (MSME). The core idea of MSME is modeling each encoding block into target template.Using Meanshift object tracking technique, the displacement between currentencoding block and the best-matched block in reference frame is directly detected,avoiding intensive searching process of the traditional methods. This cansignificant reduce the computation complexity of motion estimation module. Thelatest video coding standard H264/AVC is chosen to implement and evaluate theproposed algorithms. The original motion estimation algorithms of the referenceencoding software JM are disabled and the two proposed algorithms areintegrated into encoder. In order to evaluate the performance of the proposedalgorithms, test sequences with various motion contents and different sizes areused for encoding experiment. The experiment results show that the two proposedalgorithms can significantly reduce the computation complexity of the originalfull search algorithm, while keeping well stability with only sacrificing a bitdegradation of image quality and a little variation of bitrate of compressed codestream.