Implementation Of AVS Video Decoder Based On DM642 And Research On Optimization Methods

AVS is a digital audio-video coding standard completely exploited by our own country. It is the second generation of source coding technology standard developed with independent intellectual property rights. It has the advantages of high- performance,low computational complexity, low-cost licensing. Its implementation indicates that China is at the leading position in the fields of multimedia research in the world, and will create considerable economic and social benefits. Therefore, the study of AVS codec will be of vital significance.The purpose of this paper is to implement AVS-P2 video decoder on the platform TMS320DM642 and to optimize the decoder. Three fast algorithms is proposed in this thesis for three core modules. The decoding speed is raised by various optimization methods.Firstly, AVS-P2 video standards and the common international video compre- ssion standards are described. Then AVS-P2 codec framework , system structure and core modules are researched. TI's TMS320DM642 digital programmable universal multimedia processor is introduced on all fronts. This paper also summarizes the common optimization methods for DSP .The major work of his dissertation is to research the overall framework of AVS video decoder based on TMS320DM642 platform and the decoding optimization method(structural adjustment procedures, Cache ,QDMA, intrinstics and software pipelining), and to optimize the core modules. Three fast algorithms are proposed for three core modules. Firstly, entropy decoding: Fast stopwatch is constructed. The original complex condition judgment and skipping are replaced by the method of taking the flag bit, The complex calculation of entropy decoding is simplified; Secondly, in the process of subpixel accuracy brightness interpolation, this paper uses strategy with the combination of boundary expansion and QDMA. This method avoids frequent boundary judgement and internal and external data interchange. The decoding time is greatly reduced; Thirdly, in the process of B frame motion vector forecasting, the fast algorithm which slices the computing from the macroblock level to the frame level and uses look-up table is adopted. This algorithm reduces redundancy computation. The decoding time is saved effectilvely with negligible space, and the decoder performance is enhanced.The performance of the decoder after optimizing is tested on a number of video sequences, the result shows that this decoder achieves higher performance in both subjective and objective aspects, and average decoding speed can reach to 40 frame/s. All of the three fast algorithms proposed in this paper play good roles and raise decoding speed.