H. 264 Study On The Performance And Optimization Of Software Codecs

Entering 21 Century,the adoption of digital video has been fuelled by the development of many video coding standards in many applications, which are targeted in different application areas. There are many video protocols such as H.26x and MPEG-x. However the requirements for video transmission are increasing with the prevalence of Internet, so the video protocols are not applicable nowadays. Therefore ITU-T put forward and released a new protocol for video compression —H.264. It has many different features and its efficiency is improved through added flexibility and functionality. Firstly, the dissertation introduced the history of the multimedia technology, the basic concept and method of video compression. Then, it described the main structure and the flow of H.264 encoder/decoder. It used a lot of advanced technologies, such as the motion estimation based on multiple-modes motion vectors, intra-prediction, integral transform, loop filter. So it can achieve excellent performance, and will be used widely in future. In despite of its excellent performance, the quantity of calculation is excessive. So the decoder/encoder of H.264 runs very slowly. In general, it is hard to meet the requirement of real time applications. In this dissertation, we analysed the decoder/encoder of H.264 carefully, then adopted various ways to optimize it. In decoder: firstly, we adjusted the flow of the decode program, reduced the times of memory access; secondly, we use the MMX and SSE instruction to optimize the integral transform and sub-pixel interpolation,adjusted some simple but high frequency called functions to inlined asm function. For convenience to use, we made the decoder program to a library, and use it to make a player in windows environment. In encoder: the main methods to optimize are similar with decoder. In the part of motion estimation, we used the MMX instruction in the parallel calculation of the SAD. In the end, the dissertation compared the performance of the optimized decoder/encoder with the original. The result indicate the optimized decoder/encoder have cut down the running time while maintaining the performance.