Group Statistics And Gs Sim-based High Efficiency Intra Prediction Algorithm In AVS2 Video Coding

AVS2 video coding is the second part of the second generation Audio-video Coding Standard (AVS2), and it is mainly used in high definition or ultra high definition video.AVS2 has double coding efficiency than its first generation and it has comparable performance in comparison with HEVC, and it has special coding for monitoring the scene and 3D video. AVS2 not only adopts the traditional technology, but also introduces a lot of new technology. Meanwhile, these new technology impose lots of computational complexity for AVS2 coder, and it does't meet the needs of real-time. Therefore, we researches the core technology under the configuration of All-Intra, and propose a novel group statistics-based intra prediction mode decision algorithm and a GSSIM-based coding unit decision algorithm,which aiming at reducing computation time and improving the coding efficiency.Firstly, the thesis introduces the traditional video compression processing technology and the development of video coding standards at home and abroad. Then, the core technology of AVS2 video coding will be introduced including intra-prediction,inter-prediction, logical units, transform, quantization, loop-filter as well as entropy coding.Secondly, the thesis introduces the algorithm of intra luminance prediction in AVS2 and the detail in reference software of RD14.0. This algorithm includes two aspects: RMD process and RDO process. The RMD process is used to choose 9 candidates from all the 33 prediction modes, and those candidates will enter RDO process in order to select the optimal mode based on the rule of rate distortion optimization. From the observation, the number of candidate modes for RDO process is still high and have some possibilities to reduce. Hence,a group statistics-based fast intra luminance prediction mode decision algorithm is proposed.In the algorithm, the all 33 prediction modes can be merged into 7 groups according to similar number of prediction modes represent for a rough same direction, then considering the particularity of MPMs and the overall distributions of 9 candidate modes out of RMD process, the candidate modes for RDO computing process can be reduced efficiently, thereby reducing the computational complexity and improving the efficiency.In the aspects of coding unit decision of AVS2 video coding, a unique quad-tree reeursive division structure is adopted, and determining the optimal partition of one LCU requires 85 RDO calculations for different CU sizes, which is a time-consuming work. So, a GSSIM-based fast coding unit decision algorithm is proposed. In this algorithm, according to the spatial correlation of the image and the characteristic of gradient structural similarity(GSSIM), the GSSIM can be analyzed between the current CU and its neighboring coded CUs. Thus, based on the costs of GSSIM and the sizes of neighboring coded CUs we can early terminate the partition of current CU or skip the RDO computing of some CU sizes.Finally, the proposed algorithms were implemented on the RD14.0, and a lot of experiments and statistics have been done on different video sequences. The experiments results show that the proposed fast intra luminance prediction modes decision algorithm can decrease the video encoding time by 26.79%?32.82% and the proposed fast CU size decision algorithm can decrease the video encoding time by 26.05%?28.45%.