Study On Algorithm And Improvement Of Integer Pixel Motion Estimation In H.264/AVC

With the improvement of digital technology and network applications, and the increases of the number of mobile multimedia communications, the size of video data becomes more and more large. Because of the limited channel capacity, the date compression technology should be used before transmitting this large size of data. The video coding standards H.264/AVC provides many supports for the date compression technology, with the excellent compression performance. However, the motion estimation algorithms recommended by H.264/AVC have higher complexity and it is a serious bottleneck of achieving real-time video coding. Thus, Fast implementation of the motion estimation has the important significance for improving the performance of video coding.Aiming to the fast implementation of the motion estimation based on the video coding standards H.264/AVC, the research on two issues is carried out in this paper. One topic is focused on the study of the hybrid multi-hexagon-grid search (UMHexagonS) algorithm and comparison and analysis of various optimization strategies of the UMHexagonS algorithm, which are proposed by scholars at home and abroad. Another one is focused on the improvement of fast integer-pel motion estimation algorithm using directional information algorithm proposed in this paper, in which the validity of the proposed algorithm has been verified in software tool JM12.2.The basic principle and techniques of H.264/AVC video compression standard are described firstly in this paper, followed by the detailed introduction on some classical motion estimation algorithms, such as three-step algorithm, four-step algorithm, diamond algorithm, and hexagon algorithm, and their advantages and disadvantages. These algorithms can accelerate the speed of motion estimation search. However, they could easily become trapped in a local minimum for large size search step. The UMHexagonS algorithm adopted by the video coding standards H.264/AVC is explored in detail. Various optimization strategies proposed by scholars at home and abroad for the UMHexagonS algorithm are analyzed and compared, including initial search point prediction, matching criterion, search pattern, adaptive search range and early termination. The Multi-Octagon-Grid algorithm, which is an optimization algorithm of UMHexagonS by changing search pattern, is studied by simulation.An improved fast integer-pel motion estimation (ME) algorithm using directional information is proposed. The ME cost distribution information, obtained by horizontal-direction search and vertical-direction search in the stage of the unsymmetrical-cross search, has been explored to reduce searching points. Thus, the 25 points square search of the original UMHexagonS algorithm is adaptively adjusted to an utmost seven points search, and the 16 points multi-hexagon-grid search to an utmost 4 points search. Performance analysis for the proposed algorithm and comparisons with UMHexagonS algorithm and Multi-Octagon-Grid algorithm are conducted by using the same sofeware and the same test sequence. Experimental results show that the integer-pel ME time and search points are considerably reduced by using our method, while the rate-distortion performance and bitrate can retain almost the same.