Research On H.264 Inter-frame Prediction Algorithm

H.264 is a highly compressed digital video codec standard which was put forward by a joint video group (JVT, Joint Video Team) formed by the ITU-T Video Coding Experts Group and ISO/IEC Moving Pictures Experts Group. H.264 is characterized by better coding efficiency and network flexibility. H.264 uses a lot of new techniques including a quarter-pixel accuracy motion estimation, tree motion compensation, multiple reference frame model, integer DCT transform, entropy coding, deblocking filtering and ect., because of that the encoding efficiency of H.264 is greatly improved, however the computational complexity of encoding is several times to ten times higher than that in the past standard. This high computational complexity greatly limits the application of H.264, especially in the field of real-time video coding. So how to reduce the computational complexity of encoding without changing the high compression quality becomes a research hotspot. This article is studied for the largest operation part of inter prediction.Firstly, the basic principle of the H.264 standard has been studied and key technologies to H.264 coding standard are analyzed. The author uses the characteristics of tree motion compensation, improves the algorithm for zero-block motion decision search and no longer repeat the all-zero sub-block motion search, which reduces the number of motion estimation search point, thus lower the computational complexity of encoding.The experiments demonstrate that the algorithm meets the requirements of reducing motion search points greatly with the premise of SNR(Signal Noise Ratio) warranty.Secondly, the author studies classical inter mode selection of algorithm of H.264 and proposes a mode selection algorithm which bases on the analysis of the coded macroblock. Do simple cluster analysis of coded macroblocks around the current coding macroblock by taking advantage of correlation between macroblocks and then according to the result of the analysis choose corresponding encoding. The experimental results show that encoding time on average decreases by 20.71% to 36.37% on the basis that the average SNR loss is 0.07dB and the average code rate. rises to 1.84%. Improve clustering process by replacing the Simple clustering algorithm with Mean Shift clustering algorithm and proposes a mode selection algorithm based on Mean Shift clustering algorithm. The experimental results show that encoding time on average decreases by 35.07% to 38.05% on the basis that the average SNR loss is 0.05dB and the average code rate rises to 1.76%.