Research Of Motion Estimation Algorithm In Video Compression

Motion estimation is one of the core-techniques of video coding.Motion Estimation and motion compensation can reduce the large amount of temporal redundancy that exists between frames of video sequences,which leads to high compression.Therefore,research on efficient video compression coding technology has already been a hot topic for technicians and researchers all over the world.Motion estimation and motion compensation techniques can effectively eliminate temporal redundancy between video sequences,which have been widely applied to popular video compression coding standards to improve compression efficiency.Nowadays,it is an important issue in video coding area to develop a speedy,robust and efficient motion estimation algorithm.Among current motion estimation methods, block-matching algorithm(BMA) is the most popular one owing to its simple principle and suitability for implementation,many fast block-matching algorithms such as Three-Step Search(TSS),Diamond search (DS) have been proposed to reduce computational amount of Full Search(FS) which is not fit for real-time applications because of its unacceptable computational cost.Although these traditional fast algorithms can improve search speed significantly,they have an inherent shortcoming of being liable to get trapped in local optima.Traditional fast algorithms can degrade the quality of motion estimation to some extent.It is a problem that needs resolute urgently.In this thesis,the main task is to analyze and study several classic motion estimation algorithms, and on the basis of these algorithms,an attempt is made to implement and improve the performance of motion estimation algorithms.Firstly,this thesis elaborates the essential principle of the motion estimation based on block matching,introduces Full Search method and several typical motion estimation fast algorithms based on block matching,analyzes each characteristic,and evaluates their advantage and disadvantage through the data get from experiment.Secondly,an improved motion estimation fast algorithm is proposed—An Novel Oriented Diamond-Arc search(NODAS) is presented.Based on the cross-center-bias property and spatial-temporal correlation of the motion vector field,combined with the prediction of initial search point,and according to the prediction of unblocking,different strategies are adaptively utilized.The algorithm in this paper substantially improves the motion estimation's speed,while ensures the search accuracy.Experimental results indicates that NODAS suitable for(quasi-)stationary and large motion searches.In addition,the research of the distributed video coding also proposed in this paper. Finally,experiments are made to prove the performance of the new algorithm.Then we can find that they are better than others by analyzing the result of the experiment and the anticipated results are realized. According to the experiments data,the following conclusions can be drawn,for whatever small,middle and large motion of video sequences,the algorithm in this paper keeps the two relatively high-performance in the search and accuracy and search speed.The algorithms takes full advantage of spatial-temporal correlation of the motion vector field to predict the initial search point,it is suitable for(quasi-)stationary and large motion searches,and can avoid mistakes and improve the search efficiency.