Research On Motion Estimation Algorithms In Video Compression

A large number of researches on internet, multimedia technology and video have promoted the continuous development of video compression technology. Due to the network bandwidth limitation, real-time transmission of video information requires efficient video coding. Motion estimation can eliminate the temporal redundancy between video sequences effectively in order to improve the efficiency of video compression. As one of the key technologies in video compression, motion estimation is the most complex and time-consuming part of video compression coding system, so fast and efficient motion estimation algorithms has always been an important research subject in academia and industry.Block matching algorithms which are widely applied for its simplicity, practical and easily implemented in hardware has also been extensively studied. This dissertation firstly analyses the video coding system, summaries several main techniques in it, and then focuses on the block matching algorithm. The block matching criterions are discussed and many classic block matching algorithm are analyzed and summarized.An adaptive video motion estimation algorithm based on Bayesian decision is proposed in this dissertation to improve the speed and accuracy of the motion vector prediction in video motion estimation by the research of ARPS-3 and its feature. The algorithm makes full use of Bayesian theory, space consistency of MV and the encoded frame's influence on current frame. Based on the known video blocks of the previous frames and the current frame, the probabilities of minimum space distance among the motion vector of the target search block and the three surrounding blocks are calculated, and the MV prediction is determined according to the Bayesian decision.By analyzing the feature of motion vector, this dissertation proposes a fast algorithm based on motion classification and direction prediction in which the spatial-temporal correlations of adjacent blocks'motion vectors are used. The algorithm predicts the motion classification and relative motion of blocks and adaptively chooses different search method with motion classification information of blocks. For relative fast motion blocks, it predicts the motion direction and chooses corresponding search patterns.Experiments are made to prove the performance of the new algorithms. By Analyzing the experimental forms and simulation map data we can find that they are superior to the classic block-matching motion estimation algorithm. According to the experimental data, the following conclusions can be drawn: An adaptive video motion estimation algorithm based on Bayesian decision and a fast algorithm based on motion classification and direction prediction which proposes in this dissertation can save more search time than other traditional algorithms with the same accuracy.