H.264 Video Coding Standard Analysis And Algorithm Optimization

H.264/AVC is the newest video coding standard jointly developed by ITU and ISO/IEC. Compared with its predecessors, H.264 employs a lot of new techniques and is a more"network-friendly"video coding standard with much higher coding efficiency. However, the gain of its high performance is at a price of heavy computational complexity. Therefore, how to reduce its computational complexity while still maintaining almost the same coding efficiency has become a crucial problem in a practical H.264 codec realization. Consequently, the optimization of H.264 has become an important research topic in the image processing society and is the main concern of this dissertation.To figure out how to effectively reduce the computational complexity of H.264, H.264 is firstly introduced in detail and its key techniques are thoroughly analyzed, then the optimization algorithms for the time consuming modules, Intra Prediction, Inter Prediction, and Motion Estimation, are proposed in this dissertation.First, after the analysis of intra-prediction employed in H.264, a novel fast algorithm for intra-prediction mode decision based on pixel's gradient direction is presented. The proposed algorithm uses two candidate mode selection rules to choose the candidate modes firstly, and then employs the average SATD values of them as the threshold to further eliminate the unlikely modes. As a result, the computational load is decreased dramatically.Second, the key techniques of inter-prediction, such as the inter-prediction modes, exhaustively mode decision method and so on, are discussed. Then, a novel fast inter-prediction mode decision algorithm is proposed to speed up the encoding process based on the local macroblock motion activity. The proposed algorithm constantly starts with the SKIP mode and benefits from it as an early termination stage to possibly terminate further checking of other modes. Next, the city-block lengths of the concerned motion vectors from adjacent macroblocks will be used to estimate the current macroblock motion activity so as to reduce the number of modes needed to be checked.Finally, based on the analysis of the characteristic of motion estimation in H.264 and some observations from the experiments, a novel fast motion estimation algorithm is suggested to reduce the computational load of motion estimation dramatically. The proposed algorithm combines predictive mode reduction technique, selective pixel matching measure and simplified reference frame discrimination strategy to speed up the full-pixel motion estimation process. Then, a direction based selection rule is employed to further decrease the sub-pixel candidate search points. Experimental results have shown as follows: 1) The proposed fastintra-prediction mode decision algorithm reduces 45% computational complexity on average, while incurring only 0.040 dB loss in PSNR and 2.80% increment on the total bite rate compared with that of exhaustive mode decision, which is a default approach set in the JM reference software. 2) The proposed fast inter-prediction mode decision algorithm achieves a reduction of 56% computational complexity on average, while incurring only 0.059 dB loss in PSNR and 0.20% increment on the total bit rate compared with that of exhaustive mode decision, which is a default approach set in the JM reference software. 3) The proposed fast motion estimation algorithm achieves a reduction of 87% computational load on average, while incurring only 0.033 dB loss in PSNR and 0.56% increment on the total bit rate compared with that of exhaustive motion estimation process, which is a default approach set in the JM reference software. In summary, all the fast algorithms proposed in this dissertation can be extensively used in all the H.264/AVC video coding standard application, and also can be served as an example on the other multimedia software algorithm optimization.