Research on H.264 Inter-frame Prediction Algorithm
|School||Shenyang University of Technology|
|Course||Computer Software and Theory|
|Keywords||H.264 Motion Estimation Mode selection Inter prediction|
H.264 is a highly compressed digital video codec standard , proposed in 2002 by the composition of the ITU-T Video Coding Experts Group and the ISO / IEC Moving Picture Experts Group Joint Video Team (JVT, Joint Video Team) , which is characterized by has a better coding efficiency and network adaptability . H.264 uses including the quarter - pixel precision motion estimation, the tree motion compensation, multiple reference frame mode, integer DCT transform and entropy coding , and deblocking filtering technology so that the coding efficiency is greatly improved , but brings a huge encoding computational complexity , several times to ten times higher than the previous standard . So high computational complexity , which greatly limits the application in the field of real-time video encoding H.264 Therefore, how to reduce the computational complexity of the encoding in quality assurance compression premise degrees become the hotspot of study . H.264 operation amount of interframe prediction section . First , the basic principle of the H.264 standard , the key technology of the H.264 coding standard , such as the analysis . H.264 tree motion compensation features to improve the motion estimation algorithm of the all-zero block judgment , no longer be repeated for all - zero sub-block motion search , reducing the number of points of the motion estimation search , thereby reducing the computational complexity of the encoding . The experiments show that the algorithm is the basic guarantee for the signal-to-noise ratio under the premise to meet the requirements to reduce motion search point . Secondly classic H.264 inter mode selection algorithm research, based on the coded macroblocks analysis mode selection algorithm . Simple clustering analysis using the correlation between the macroblock and the macroblock has already been encoded on the current coded macro - blocks surrounding the motion vector of the macro block , selected according to the analysis of the results after the corresponding coding mode . The experimental results show that the average loss in the signal-to-noise ratio of 0.07dB, bit rate average rose to 1.84% based on the coding time dropped by an average of 20.71% -36.37 % . Then on the the encoded macroblock analysis mode selection algorithm clustering process improvement, instead of a simple clustering algorithm using Mean Shift clustering algorithm , and propose a mode selection algorithm based on Mean Shift clustering analysis , experiments show that in the signal-to-noise than the average loss of 0.05dB , bit rate rose to an average of 1.76% based on the encoding time dropped by an average of 35.07% -38.05 % .