| With the rapid development of the Internet and mobile communication technology, the demand for video coding becomes even more urgent than before. In order to break through the dilemma which lies in the long-term monopoly of the foreign standards, the AVS standard has been proposed in China. The standard preliminarily reaches the advanced technical level. At present, the open-source XAVS encoder appears to be incapable for the encoding of HD video. In this paper, we optimize XAVS encoder from various perspectives so as to improve the performance without a remarkable decline in video quality, thus making it possible for the encoder to support the HD encoding.Firstly, the paper introduces the theoretical study of the AVS standard and emphasizes on the similarities and differences in comparison to H.264 standard. Meanwhile, a brief introduction to the AVS encoder is presented in terms of modules, which lay theoretical foundation for the encoder optimization. To make full use of the function of the multi-core processor in video encoding, this paper introduces the concept of parallel programming. In addition, the acceleration performance of multi-core parallel is analyzed by the Amdahl’s law.The third part of this paper gives a deep optimization on the XAVS encoder, focusing on improving the encoding frame rate. Since the mode decision module accounts for about 70% in the encoding process, and the motion estimation is the most time-consuming module, this paper firstly optimizes the UMHS motion search algorithm from many aspects so that the search points and steps can be reduced significantly. Upon testing on many video sequences, we can conclude that the optimized algorithm achieves an improvement in frame rate by 12% to 19%, while the loss of PSNR (Peak Signal to Noise Ratio) less than 0.1dB and the rise of bitrates no more than 0.5%. Furthermore, since the early decision for SKIP mode is of great significance, this paper introduces a SKIP decision algorithm based on HVS (Human Visual System). By modeling the luminance, contrast ratio, variance and other parameters of the macroblock, this paper uses a new model of SKIP early decision to avoid the complex steps of motion search. The experimental results show that the improvement of frame rate is about 9% due to the 57% rise of the proportion of SKIP mode. The optimization of motion estimation and SKIP decision method are merged into mode decision module, therefore the frame rate increases at about 21%.The fourth part of this paper has put forward multi-thread parallel programming in the AVS encoder. Through the analysis of the parallel acceleration performance of the encoder, we find that there exists a widespread problem of uneven distribution of computation in the parallel encoding process. This paper predicts the coding mode in advance so that each processor core allocates roughly similar amount of computation, thus achieving load balancing and improving the parallel performance. The experiments which are performed on many processors reveal that compared with the conventional parallel coding, the frame rate of XAVS encoder increase by about 20%, the parallel acceleration ratio can be more than 3.6 times, while the rise of bitrates is within 1% and the loss of PSNR is no more than 0.35dB. |
PDF Full Text Request