Mode Decision And Fast Search Algorithms Optimizing In H.264 Video Coding Standard

New video codec standard H. 264 presented by JVT is advanced than other early same standards and with excellent performance than MPEG4/H. 263 ++. H. 264 strengthens adaption for network and accelerates research of video compression algorithm. In whole process of video coding, motion estimation and motion compensation which are bottleneck of video codec spend much more time than others. A number of efforts have been made to explore integrate pixel search algorithm which based on block-matched, and get achievement in some degree. As for sub-pixel research and mode decision algorithm should be further researched. For H.264 provides more prediction modes than early standards, mutiple reference frames and block sizes, research of mode decision and fast search algorithms is significant for practice of video codec.In this paper, principle of video coding and improved performance of H. 264 are introduced. Mode decision for intra prediction algorithm is inhanced, and mode decision of inter prediction and fast sub-pixel search in motion estimation algorithm is validated.Adopted two-level select intra prediction mode decision algorithm employs computation of RD function and SAD to select most possible modes, sets threshold to terminate algorithm earlier, then reduces complexicty largely. In first step, filter out least probable modes, only consider one quarter of all pixels when compute SAD or SAE, for this method can reduce computation load and is almost same accurate.Combination of intra prediction mode decision and fast search algorithms can reduce computation load by searching only candidate modes and optimizing sub-pixel search algorithms. So it can improve efficiency of video coding significantly. Modes which are removed in first step are not searched and computed to reduce computation load. Inpart of mode decision, transform block to transform domain then judging by SATD which are different from large computation in spacial domain. When compute sub-pixel by linear interpolation, algorithms adopt both pre-save computed result and immediate computing, get outstanding performance, but should be improved in further practice.