Implementation And Improvement Of H.264 Encoder Based On DSP

As a latest international video coding standard, H.264 adopted many new technologies to improve coding efficiency compared with previous standards. Therefore, H.264 has been applied in more fields. However, the high coding efficiency results in high computational complexity, which restricts the application of real-time scene. DSP is famous for its high computing-speed and computing capability. The high-performance digital media programmable processor, TMS320DM642, manufactured by Texas Instrument, has powerful ability to data and much peripheral equipment for digital media, which will develop H.264 encoder well.The work of this paper is focus on H.264 algorithm and the optimization of the H.264 encoder based on DSP.For H.264 algorithm, firstly, we analyze the process of the Intra and inter prediction and some existed solution at present. According to the statistical character of the selected coding modes, we proposed a new mode decision algorithm for coding modes based on the activity of macro-blocks and the priority sequence of modes. In order to eke disadvantage of video quality which is possibly caused in the process of mode decision out, a 1/8 pixel motion estimation algorithm based on mathematical model is proposed.For the optimization based on DSP, 2 levels optimization are performance. Firstly, optimizing the compiler options and the storage system called global optimization. Secondly, namely partial optimization, we optimize C code according to the structure of DM642 and the algorithm of H.264 encoder.After the improvement and optimization on encoder, the performance of encoder is improved obviously. Experiments results show that the coding time reduces about 30% without obvious PSNR change after the improvement of mode decision. Optimization on DSP platform also achieves a lot. Compared with non-optimization, the coding time reduces about 80%.