Research On Motion Estimation In Video Coding Standard Of AVS

To save the network bandwidth,storage space,video coding has become one of the hot issues of research and industrial applications.So far the international community has set a series of video coding standards,such as MPEG-x,H.26x.To improve the video technology in China, AVS video standard is developed by the Audio Video Coding Standard Working Group of China, which was approved in June 2003.AVS video standard provides a good trade-off solution between complexity and encoding efficiency.In video coding system,motion estimation plays a very important role in eliminating inter-frame redundancy and improving the performance of video coder.On one hand,the accuracy of motion estimation affects the efficiency of the video coder.On the other hand,the complexity of the encoder lies on that of the motion estimation algorithm.This paper studies the motion estimation with high complexity in AVS,including the integer pixel part and the sub-pixel part.For the integer pixel part of UMHexagonS (Unsymmetrical-cross Multi-Hexagon-grid Search),the original algorithm continues the hexagon search or diamond search after the initial search point prediction with good effect.That produces some redundant search,which is the same as some search pattern is fixed without considering different cases.On the base of that,an early termination is added after the initial search point prediction and the search pattern is adapted with the different block type.For the sub-pixel part,the original algorithm uses the HFPS(Hierarchical Fractional Pel Search) algorithm if the block type is 16×16,16×8 and 8×16.HFPS algorithm is a kind of full search algorithm of sub-pixel with the best performance,but the most time-consuming.Considering that the motion vectors of sub-pixel is usually around the initial search point,an algorithm with adaptive search pattern based on diamond pattern is proposed in this paper.It decides the next step after judging the best matching point's position.The test shows that the improved algorithm could reduce the time-consuming in motion estimation with only a little decrease in bit rate and hardly loss in image quality.It basically achieves the anticipative goal.