Research And Optimization On Motion Estimation Algorithm In H.264

The characteristics of H.264 video compression standard are low bit rate, high quality and high compression ratio. However, these performances are at the expense of increasing computational complexity. Block-based motion estimation and motion compensation are critical components in video coding systems. Multiple coding modes and multiple reference frames are used in H.264, which make motion estimation become the most time-consuming part in H.264 encoder, for it occupies more than 60% of the total computation. As this reason, how to reduce the motion estimation time under the premise of ensuring high-performance is one of hotspot in current video compression research.In this paper, firstly, the basics of video encoding, various kinds of international video compression coding standards are expounded. Secondly, the key technologies in H.264 are introduced respectively. Thirdly, the theory of motion estimation and several efficient fixed pattern motion estimation algorithms are introduced principally. And then, on this basis, the characteristic and strategies of H.264 motion estimation are summarized and introduced simply. Finally, an improved motion estimation algorithm is proposed.A large number of algorithm researches reduce computation by reducing local search points. In this regard, the fixed pattern motion estimation (FPME) has got great achievements. Whichever the fast fixed pattern search algorithm is, it does not use image feature. This kind of fast algorithm need search all of the candidates exhaustively, and the search would easily be trapped into a local minimum.In video sequence, people are often interested in moving objects. There are many static blocks in background, and the movement between the two different backgrounds of inter-frames is often considered as a result of camera moving. Object motion vector will not depart from the starting point very far. Edge information is often lost due to the low matching. This paper describes the characteristics of motion vector field, and analyzes the best mode statistical law of inter-frame video coding. To overcome the weakness of FPME, which is not use the feature of image, a variable block-size motion estimation algorithm based on block feature classification (VBBFC) is presented. Skip mode and stationary 16×16 block are judged at first. After that, coding blocks are divided into 3 movement types: low movement block, medium movement block and high movement block, which is based on the prediction by using the motion vectors of adjacent blocks. At last, the appropriate possible coding modes are selected according to different movement type. The proposed algorithm is based on these search strategies, including static block determination, background-block, foreground-block and edge-block detection, start point prediction, the improvement of search window and accurate edge block matching.Comparing with full search algorithm (FS), UMHEX and SUMHEX, EPZS, experimental results reveal that the proposed algorithm can speed up motion estimation greatly, while the PSNR and bite rate almost don't increase at all.