Research On Key Algorithms Of Mobile Video Coding Standard AVS-M

With the emergence of communication networks, along with the rapid growth of multimedia service, the role played by multimedia communication has become increasingly important and essential in next generation communication system. Therein, video coding is the fundamental and kernel technique in emerging multimedia/video communication. AVS-M is the latest video coding standard developed by China for mobile application, which is the part 7 of domestic AVS (Audio Video coding Standard) series standard. AVS-M is designed to give a technical solution for a broad range of applications, such as interactive storage media, broadband video service, multi-media mail, long distance video surveillance, video conference, and video phone, etc. It is the video coding standard orienting next generation mobile communication in China.The design of AVS-M is based on conventional block-based motion-compensation hybrid video coding concepts. For the enhancement of the coding efficiency, AVS-M adopts new tools as followed: new intra frame prediction, variable block-size with seven block sizes in motion prediction, multiple reference picture, quarter-pixel accuracy for motion vectors, integer transform and loop deblocking filter. With these features, AVS-M can provide improved performance close to the newest international video coding standard H.264/AVC. At the same time, the complexity of the AVS-M codec is increased remarkably. Accordingly, the real-time video codec demands higher-powered processor and faster-speed algorithms. So the main endeavor of this paper is to achieve maximal processing speed and better video compression quality through optimization for intra prediction, inter prediction, motion estimation, integer transform in the AVS-M encoding process. The major works of the thesis are as follows.1. To reduce the complexity of intra prediction, a fast intra prediction mode selection algorithm is proposed. It uses the information of edge direction in the current coding macroblock and spatio-temporal correlation of intra prediction modes to fast determine the best mode with the RDO (Rate Distortion Optimization) criterion.2. To reduce the complexity of inter prediction, a fast inter prediction mode decision algorithm is proposed. It classifies all modes by both homogeneity and spatio-temporal correlation of macroblock based on the statistic analysis of the results of the optimal modes. The SKIP mode can be early terminated and other modes can be compared type by type. So the number of calculated modes can be decreased efficiently.3. As the most important module in the AVS-M encoder, motion estimation is accelerated in two perspectives. In the process of integer pixel motion vector search, by calculating and analyzing the relevant motion vectors of current block, an adaptive fast motion estimation algorithm is proposed. The algorithm can effectively reduce computation in the optimal integer pixel motion vector search process by some effective techniques, which are composed of the prediction of initial search point, half-stop criteria and adaptive search modes. Meanwhile the algorithm can maintain comparable search quality. To reduce the complexity of fractional pixel motion estimation, a prediction based fractional pixel motion estimation algorithm is proposed to replace the full search algorithm. It can efficiently decrease the number of search points and come to a good matching.4. For most sequences at middle or low bit rates, the values of prediction residuals usually are very little, and most of them will be reduced to zero after transformation and quantization. All-zero block detection is one of the common optimization methods in low bit rate video coding. According to the new characteristics of AVS-M standard, the threshold for all-zero block detection is derived and an advanced stop principle of motion estimation based on all-zero block detection is proposed. In accordance with the feature of multiple encoding modes, all-zero block detection is applied to the AVS-M mode selection, and further improves the efficiency of motion estimation. Experimental results show that the proposed algorithm can reduce encoder complexity significantly while maintaining PSNR and bit rate almost unchanged.The conclusions and directions for future research work are discussed in the last chapter of this thesis.