Dsp-based H.264 Motion Estimation Algorithm And Implementation

With the rapid growth of mobile communications and Internet, voice and data communication cannot meet people's need any more. The requirement for businesses of digital video and image is increasing enormously. On the other hand, video communications require very much bandwidth in comparison with other information types such as text and data. Video source must be compressed extremely so that it can be used in actual application such as storage.H.264 is the next generation video coding standard of ITU-T and ISO/IEC. Techniques the new standard uses include integer transform instead of DCT, more types of block and more accurate motion compensation for inter prediction, more modes for intra prediction, and Context-Adaptive Entropy coding (CAVLC or CABAC) and so on. In terms of coding efficiency, the new standard is expected to provide at least 2x compression improvement over previous standards.This paper introduces some algorithm optimization on several parts of H.264, and do some optimization on ADSP-BF533 EZKIT. The thesis analyze integer pixel motion search algorithm inter-prediction part which costs the most time in H.264 standard, produce our new search algorithm——NMVFAST after reading a lot of correlative research productions of national and international. The result indicates that the NMVFAST can expedite compute speed, at the same time, there isn't obvious effect for image quality after processing.In the last, the thesis discusses how to optimize the source program on ADSP-BF533 so that it can be executed in embedded system. We use mixed programming of C and assembly language and merge their merits to realize higher efficiency and shorter execution time.