Block Matching Motion Estimation For H.264 Based On Video Sequences Feature

Block-matching Motion Estimation (BME) takes more than half of computation in H.264 video coding. The matching speed and accuracy of BME directly affect the coding speed and quality. Generally, search pattern and search strategy play an important role in BME. However, the previous BME algorithms don't make use of the video sequences feature fully. Most of them adopt regular search pattern or single search strategy and can not deal with the blocks which have different motion degree respectively. So it makes the BME be deficient in adaptability and restricts the efficiency of coding to some extent.Due to the defects of previous BME algorithms and for the purpose of improving the performance of BME, this thesis proposes a Mixed Pattern Based Search(MPBS) algorithm, based on the research of video sequences feature, and after the research of how to choose the initial search point, the thesis makes a further improve for MPBS.The thesis firstly makes a detailed discussion and a deeply analysis for classic BME algorithms, and summarizes the advantages and shortcomings of those algorithms. Then through the research of video sequences feature, the thesis analyzes the center-biased characteristic of the Motion Vector (MV) and the spatial correlation of neighboring blocks. At the same time, through experimental analysis, this thesis obtains the distributed model of MV outside the initial search pattern, and finds that there are about 56.8% of motion vectors located between Minimum Block Distortion (MBD) point and Secondary Minimum Block Distortion (SMBD) point. Taking advantage of those features and the experimental results, this thesis designs a mixed search pattern and its search strategy correspondingly, including the initial crossed-pattern, the following line-pattern, and the mixed searching strategy of two patterns. Then the thesis proposes the fast BME algorithm MPBS based on the research of video sequences feature. Lastly, the thesis makes a further research on the application to predict initial search center for MPBS, and improves the MPBS by using this technique.Experimental results show that the proposed algorithm can reduce more than 50% of search points, compared with those classic BME algorithms, while maintaining the quality of pictures. When the MBPS uses the initial search center prediction, it can reduce 17.5% of search points more than original, and also improve the PSNR by 0.34dB.