Optimization Of H.264 Video Coding Algorithm And Data Memory Organization For Decoder

H.264/AVC is the video coding standard developed by ITU-T Video Coding Experts Group and the ISO/IEC Moving Picture Experts Group, which achieves enhanced compression performance and provides a "network-friendly" video representation. However, these advantages are obtained at the cost of increased computational complexity, which has more requirements on the storage and processing capability of the hardware.Motion compensation is a key component of a hybrid video coding standard such as H.264 and it can effectively remove the redundance of successive video signals. As motion estimation at quarter-pixel accuracy with variable block sizes and multiple reference frames is adopted, the computational complexity of motion estimation is increased largely and it can consume 60% (1 reference frame) to 80% (5 reference frames) of the total encoding time of H.264 codec. In order to reduce the computational complexity of H.264 codec, this paper focuses on the research work of algorithms on fast motion estimation, and new algorithms on fast motion vector estimation and multi-frame prediction are proposed respectively. They are line prediction based pseudo-diamond search (LPPDS) algorithm and spatial prediction based multi-frame selection algorithm.Small diamond search pattern (SDSP) is adopted in the line prediction based pseudo-diamond search algorithm. The algorithm first chooses the prediction point with the munium distortion as the initial search point to apply one step of SDSP search, then performs the line prediction to find the best center points of each step of diamond search, so that the best intetger position can be found in fewer steps. At the same time, the new algorithm can be used with many fractional pixel search algorithms well to reduce the computational load of the whole motion estimation. Simulation results demonstrate that the proposed LPPDS algorithm provided a search points reduction in average of about 50% compared with diamond search algorithm while keeping almost the same compression ratio and picture quality.In the spatial prediction based multi-frame selection algorithm, best reference frames of the neighbouring blocks are used to predict the best reference frame for current block, these frames are considered as the reference frame candidates of current block. The best reference frame is only selected from the candidates to reduce the computational load of multi-frame selection. At the same time an adaptive threshold for cost of motion estimation is adopted to reduce the probability that the selected reference frame is not the best. As a result the compression ratio and picture quality will be ensured.In this paper we also study the memory organization of H.264 decoder based on DSP. Memory organization is very important for the decoder software implementation based on DSP.Good memory organization will speed up data accessing and make the transplant of code on diferent plateforms easy. According to the storage capacity and the efficiency of data transfer between internal and external memory, double buffering technique is used in the memory organization of H.264 decoder. Simulation results demonstrate that the memory organization is reasonable.