Research On The Fast PU Mode Selection Algorithm Based On Consistency In HEVC Encoder

As the next generation video coding standard, HEVC was jointly developed by ITU-T and ISO, and was officially launched in 2013. The HEVC encoder reserves the traditional hybrid video coding framework, and further introduces some new coding techniques. Compared with H.264 / AVC, HEVC can reduce the bit rate of about 50% under the same video quality. However, the computational complexity of HEVC encoder is significantly increased at the same time. Therefore, reducing the computational complexity of HEVC encoder and improving the encoding speed has become a hot research topic recently. Thus, fast mode selection algorithm is studied to save coding time in this paper.In order to reduce the computational complexity of HEVC codec, various key techniques are deeply studied, especially the CTU mode selection module. It is found that the excessive PU partition modes mainly lead to the high complexity of CTU mode selection process. Accordingly, a fast PU mode selection algorithm framework based on pre-process is proposed. The proposed algorithm obtains and analyzes the main factor influencing the PU mode selection before the rate-distortion optimization of PU, and then the top several PU modes are selected as candidates. The proposed algorithm framework consists of four parts: pre-processing, consistency analysis, candidate PU mode selection, and simplified PU mode selection.Firstly, the PU mode selection algorithm is optimized in terms of the consistency of MV trend. It is found that the size of the best PU mode is bigger when the consistency of MV trend is stronger. Thus, according to the framework of fast PU mode selection algorithm, a fast PU mode selection algorithm based on the consistency of MV trend is proposed. In the pre-processing stage, a nine-points motion estimation method is presented to obtain the motion information of CUs. In the consistency analysis stage, the strength of the consistency of MV trend is determined according to the motion information obtained in the pre-processing stage. In the candidate PU mode selection stage, a few PU modes are selected as candidate modes for each CU in terms of the consistency of MV trend. Experiment results show that the proposed algorithm can select the best PU modes for some CUs rightly, and reduce the encoding time by about 16.5%?Secondly, the fast PU mode selection algorithm is improved based on the consistency of MV degree and the consistency of texture complexity. By analyzing the relationship between the best PU mode and the consistency of MV degree as well as the relationship between the best PU mode and the consistency of texture complexity, it is revealed that the stronger the consistency of MV degree and the consistency of texture complexity is, the more likely specific PU mode is selected. Therefore,a fast PU mode selection algorithm based on the consistency of MV degree and texture complexity is proposed under the framework of the fast PU mode selection algorithm. In the pre-process stage, the nine-points motion estimation method is still used. In the consistency analysis stage, the consistency of a CU is estimated based on the consistency of temporal MV degree and the consistency of texture flatness. In the candidate PU mode selection stage, motion consistency and texture flatness consistency are jointly used to select candidate PU modes for CUs. Experimental results indicate that the fast PU mode selection algorithm based on consistency can reduce the encode time by about 80%, without apparently decrease in coding performance.