Research On Optimization Of H.264 Motion Estimation Algorithm

H.264/AVC is the new generation video coding standard jointly developed by ISO/IEC and ITU-T with higher compression ratio, higher image quality and more excellent network compatibility. However, the computational complexity of H.264/AVC is much higher than the previous standard, which hinders its real-time application and generalization. In order to reduce the coding time and improve the coding efficiency of the system, it is necessary to optimize the time-consuming and resource-intensive motion estimation module. In addition, more and more attention is paid to the technology of video coding implemented on embedded system platform. And most of the related research works are focused on the software encoder transplantation to DSP platform owing to its inherent high-speed data processing ability.In this paper, the optimization of H.264/AVC motion estimation algorithm is researched, and the x264 encoder has been transplanted and optimized on the DM642 hardware platform. The main research work is as follows:(1) The related works of video coding technology and motion estimation are analyzed. And the encoding and decoding structure of H.264/AVC video coding standard is discussed along with the major coding techniques. Then the principle of motion estimation algorithm and the related techniques are presented. Moreover, several existing classical Block-Matching Motion Estimation algorithms as well as their advantages and disadvantages are analyzed.(2) In order to reduce the complexity of motion estimation and improve the efficiency of the video encoder, a multi-level adaptive fast integer motion estimation algorithm is proposed. Firstly, the frame-level search range and macroblock-level search range are predicted based on the frame-level motion intensity of the previous frame and the predictive motion vector of the current block respectively. And the minimum of them is taken as the final search range of motion estimation. Secondly, the macroblock is divided into three motion types, namely low motion types, medium motion types and high motion types, according to the macroblock-level motion intensity, and algorithmic templates of Diamond, modified Hexagon, or modified UMHexagon S are selected adaptively. Then the search strategy of direction adaption is applied to the motion estimation of the macroblocks with high motion type based on the directions of their motion vectors. Finally, the UMHexagon S algorithm is modified to improve its efficiency. The experiment result shows that the proposed algorithm can reduce the computation time of motion estimation significantly with negligible performance degradation, and it is suitable for various types of video sequences.(3) The DSP development board QXD-DM642 DVS is selected as the hardware platform, and the x264 baseline profile encoder is transplanted into DSP. Then a series of optimization measures are taken according to the characteristics of DM642 chip. Finally, the MA_DHUS algorithm proposed in this paper is integrated into the x264 encoder on the DM642 platform. The testing results demonstrate that the encoder speed up significantly after optimization.