Research On Fast Motion Estimation Algorithm And DSP Realization Of Encoder For H.264

Due to the weight of encoding time for motion search in H.264 video coding standards is great, the efficiency of whole video encoding is directly determined by the motion search algorithm. So, the study and improvement on these search algorithms are quite necessary. Further, the calculated load of coding program for H.264 standard is also very great. It causes the normal microcomputer can not satisfy the huge calculation. Therefore, the research on realization of real time H.264 video encoding by DSP has important practical meaning.Base on the deeply study of key techniques in H.264, this paper mainly discusses the improvement of motion estimation, fast multi-reference frames selection and the realization and optimization of H.264 encoder by DSP.In EPZS algorithm for video motion search, it has a lot of initial predictive vectors which causes the lowly search efficiency. We make use of the features and relationships for the vectors of space domain and time domain, to determine the initial predictive vector, and improve the search speed. In order to avoid the problem of local optimization for residual error when motion estimate in H.264 encoding, we present two fast algorithms, 4-T fast algorithm and a table-search based SAE reuse fast algorithm. On the improvement of algorithm for fast selecting multi-reference frames, a low-complexity and fast multi-frame selection method is proposed in order to avoid searching the redundant reference frames. Finally, this paper introduces the realization of H.264 encoder by DSP. It mainly discusses how to develop and optimize the H.264"baseline"encoder ported from VC platform on the hardware platform based on TMS320DM642. For the implementation of H.264 video encoder based on DSP, we should consider the characters of DSP, except for characters of video encoding algorithm. The paper is to improve the kernel algorithm efficiency of the encoder using linear assembly language and assembly language which can improve the parallelism inherent in the H.264 encoder. Combine with the above table-search based SAE reuse fast algorithm, we also make full use of two levels Cache of DM642 to optimize the encoding program.The experimental results show that the improved EPZS achieves good search performance, maintains the SNR almost the same as the full search; The 4-T fast algorithm and the table-search based SAE reuse fast algorithm also can increase the search speed with no PSNR dropping; Although the proposed low-complexity and fast multi-frame selection method may cause the SNR of peak value to be slight higher, its selection speed is over two times faster than that of the original scheme adopted in JVT reference software.