Research Of Fast Coding Unit Size Decision Algorithm Based On 3D-HEVC

With the development of technology, 3D videos?3D films and other three-dimensional technology have gradually entered and affected people's lives. Research on video compression is also becoming a hot research field. High Efficiency Video Coding (HEVC) is the latest video coding standard. As the extension of HEVC, 3D-HEVC has adopted multi-view and depth (MVD) coded format. It has proposed many new technologies which are more suitable for multi-view coding based on the coding framework of HEVC. However, no matter HEVC or 3D-HEVC, if they reduce the code rate and improve the quality of coding, they must generate high complexity and long encoding time. Therefore, it is very important to refine the encoding standard based on these two aspects.This paper introduces the research status of HEVC and 3D-HEVC firstly. Then it describes the key technologies of their encoding framework and related technologies. In the process of 3D-HEVC encoding, the recursive quad-tree structure of coding unit (CU) is significantly related to the encoding time. So reducing the time of CU division is an efficient method to improve the encoding. This paper proposes improvements on the process of CU division of texture map and depth map respectively in 3D-HEVC.In the process of existing texture map fast CU division algorithm, it is usually to set fixed threshold to early terminate CU recursive division according to features of CU. This method can save encoding time efficiently, however, it can't produce good results for all video sequences. This paper designs a Probabilistic Neural Network classifier based on intelligent algorithm. It makes the features of CU as the input of the classifier and the dividing or not as the output of the classifier. The classifier is trained by large amounts of data about CU division and tested by test data. It can produce good results for different video sequences. The simulation results show that the algorithm in this paper can guarantee video quality and. save about 43% encoding time averagely.For the research of fast CU division in depth map, this paper proposes a fast algorithm by statistical information and Bayesian Decision Theory. The algorithm reaches a conclusion by analysing the depth distribution of CU division in large Coding Unit (LCU): when all depth of CU in LCU for texture map is 0, all depth of CU in same position LCU for depth map is 0 too; when all depth of CU in LCU for texture map is 1, the depth of CU in same position LCU for depth map can be identified by threshold. The algorithm in this paper gets the threshold by the minimum error rate of Bayesian decision. When the RDCost of LCU which have same position with LCU that all depth division is 1 in texture map in depth is less than the threshold, all depth of CU in the LCU is set to 1, otherwise, the LCU is divided by quad-tree structure. Experimental results show that the algorithm proposed in this paper can save about 24% encoding time of depth map averagely on the basis of the guaranteed PSNR and bitrate.