Study And Optimization On Motion Estimation In H.264

With the continuous development of video coding technology, H.264 becomes a new standard of video-compression, which can provide higher compressibility and more friendly net interface. However, it also leads to higher computational complexity at the encoder. This is mainly due to the fact that the H.264 encoder employs more complicated approaches and features to efficiently improve the coding performance. At the same time, the encoder complexity is tremendously increased with these new approaches. Hence, it is an essential research topic to reduce the high encoding complexity while maintaining the good coding performance, especially when it comes to real-time video coding applications.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. The research on looking for a motion estimation algorithm that can get an effective and accurate motion vector quickly becomes a hot topic at present.By studying the existing rapid motion estimation algorithms, we find that most of the existing fast motion search algorithm seldom examine all possible candidates within a search area, they may fail to find the optimum point for sequences having fast and/or random motions. To alleviate this problem, we propose a measure estimating the distance between the current search point and the optimal point, which combined with hierarchical searches. The proposed algorithm improves search speed of sequences with fast and/or random motions as well as PSNR performance. Experimental results and comparative analysis are given to demonstrate that our proposed algorithms can effectively fiter and exclude some unlikely candidate modes and can achieve a fast speed-up factor compared with the current Fast Full Search algorithm in JM8.5. Moreover, our algorithms can effectively control the PSNR loss.Also, a simple algorithm (HCS) of fixed researching mode is designed by the thorough study on the existing calculate way. HCS uses a Large Cross Search Pattern (LCSP), which a Pixels from the center to the four point around. And HCS doesn't limit the times of small cross search pattern matching. The Performance of HCS is better than that of HEXBS. Based on HCS, PHCS is designed by making use of prediction, which is a simple improvement, but its function has already can place on equal footing with that of MVFAST.