AVS Coding Algorithms Researching And Its Application On DSP

With the development of wireless network and audio and video technology, the application of mobile multimedia plays an important role in social life. Data compression is very important in the application of multimedia. AVS, Audio and Video Standard, established and owned by China, is a digital audio and video coding and decoding standard. Its applications in the area of network streaming media, IPTV and video surveillance are an increasing concern. One key for broad application of AVS is to implement efficient and rapid software AVS encoder.This paper is to optimize the AVS encoder mainly in the aspects of algorithm implementation, code implementation and the processor characteristics. It does in-depth analysis and study of motion vector search algorithm and macro-block mode decision algorithm in the part of inter mode prediction.1. Improved the classic diamond search algorithm of motion vector search. The idea is to change the search mode and do the motion vector search using the large diamond pattern and small farmland pattern. It reduces the total search computational complexity by 14 percents average for exquisite movement image sequence while maintaining the same performance.2. An adapt threshold fast mode decision algorithm is presented. It reduces computational complexity about 70 percents by comparing the mode cost and selecting the best mode to encode while keeping the same performance with origin algorithm.3. A fast inter prediction mode decision algorithm is presented. Aimed to the character of surveillance video, it adopts down sampling video pre-processing and early termination rules, reduces about 50% computational complexity in the process of video coding and 20% bit-rate while maintaining the same subjective performance.At last, efficient rapid AVS encoder on TI TMS320DM642 DSP Platform is implemented by optimizing AVS encoding algorithm, designing data structure and programming frame and optimizing code. It can do encoding at speed of 25 fps for the images whose size is D1 and has been used in the network video surveillance.