| Audio and Video coding Standard(AVS) is China's second generation source coding standard developed on our own, of which we have complete intellectual property rights. Compared to other popular audio and vide coding standards around the world, AVS has many advantages, such as high efficiency, low computational complexity, low charge for property, etc. The implementation of it would make great importance to the development of our multimedia industry and create considerable economic and social benefits.This standard is composed of 9 parts, of which part2 and part7 are video coding standards.So far the referenced software of video coding and decoding for enterprise has been released by the AVS working group. In order to realize the compression high efficiently, AVS part2 (AVS-P2) adopts a lot of new technologies like H.264/AVC did, such as intra prediction, motion estimation of flexible block size, multi reference frames, motion vector with precision of 1/4 pixel,integer cosine transform, context-based 2D-VLC and Exp-Golomb coding, loop filter and so on. Whit the adoption of these technologies, the compression efficiency and decoding quality has been improved greatly.Meanwhile these new coding technologies also bring in the complexity in basic modules and algorithms, which slows down the coding speed and makes it hard to use practically, especially in those areas requesting real time. The purpose of this thesis is to optimize the AVS-P2 encoder and research the fast algorithms for core modules, in order to improve the coding speed and lay the foundation for its implementation on DSP platform.The main work of this thesis includes:First, it introduces the prevalent international video coding standards briefly, and makes a compare between the AVS and H.264/AVC.Also the architecture and the coding flow of the AVS encoder is investigated, together with some important compression technologies such as intra prediction, inter prediction, transform and quantization, entropy coding and loop filter.After deep investigation of the main entropy coding technologies such as context-based 2D-VLC and k order Exp-Golomb coding, we make an optimization on the storage architecture of 19 VLC tables used in 2D-VLC and modify the table switching methods, which save about 86% memories of the VLC tables. Also through lots of correlative experiments, we obtain the distribution of the symbol looked up from VLC tables. According to the distribution, new k order Exp-Golomb coding method is adopted to reduce the encoding time.Finally we place our emphasis on the research of the algorithms and programs in motion estimation of AVS, especially in the algorithms of fast sub-pixel motion search. According to the video sequences with different motion types, the distribution rate of the best matching positions is obtained by experiments and statistic analysis. Based on such distribution rates, one new sub-pixel searching strategy based on block-motion and a new Big-Small Diamond Searching (BSDS) method suitable for large-motion blocks are proposed. And the combination of this strategy and BSDS leads to an adaptive sub-pixel motion search algorithm for all block-motion types. The experimental results prove that such new combined searching algorithm can reduce the total encoding time by 10%-20%, with the picture quality guaranteed. |
PDF Full Text Request