Research On Fast Mode Decision Algorithm Of 3D-HEVC Intra Coding For Depth Image

With the rapid development of the film industry and emerging intelligent visual interaction technology,3D video has become a research hotspot in recent years.The two major organizations of international video coding standards,ITU-T and ISO/IET,jointly developed a new generation of high-efficiency three-dimensional video coding standards(3D-HEVC).In order to significantly reduce the number of viewpoints,3D-HEVC significantly increases the depth image of the video scene geometry,which significantly improves the 3D video compression efficiency,but the computational complexity of depth image coding is very high,and the encoding time is about six times that of the color image.Therefore,in order to reduce the computational complexity of 3D-HEVC depth image coding,this paper studies the decision-making part with higher computational complexity,including CTU partitioning,intra prediction mode selection and depth prediction mode selection.The main work done in this paper is as follows:For high-resolution video,our thesis focuses on the research of mode decision,which consists of the prediction of the CTU depth,the selection of intra prediction modes and the selection of inter prediction modes.The main works and achievements of our thesis are as follows:(1)For the CTU partitioning process,a fast algorithm for depth image coding unit partitioning based on texture feature analysis is proposed.The primary texture feature analysis is carried out by using the correspondence between the depth image texture variation feature and the CU partition feature.The fine texture feature analysis is performed according to the CU internal pixel distribution statistical characteristics.At the same time,the CTU is used to predict the depth range of the CTU and terminate the part in advance.In addition,this paper makes use of the test sequence texture feature and the gray level number and CU texture complexity threshold in the QP value adjustment algorithm to make the algorithm have better adaptability.(2)For the intra prediction mode decision process,a fast selection algorithm based on machine learning for intra prediction mode is proposed.The random forest prediction model is obtained by setting the category label to 37 intra prediction modes,extracting relevant features and performing offline training,and using the model to predict the current PU best category label,and establishing a two-level prediction mode list reduction mode decision process corresponding to the category label.The number of prediction modes for Hadard code optimization.At the same time,the number of rate distortion candidate mode lists is reduced by using the Hadamard cost difference magnitude relationship between the best candidate modes.(3)For the computational complexity introduced by depth image prediction mode DMM,a DMM prediction mode decision algorithm based on pixel feature analysis is proposed.Firstly,the distribution characteristics of the current PU internal pixels are combined and the correlation between the current PU size and DMM mode selection is combined.The DMM feature identifier of the current PU up and down and left and right is calculated to determine whether the current PU is suitable for adopting the DMM prediction mode,and the DMM search process that is not suitable for the DMM prediction mode PU is skipped,and the decision process of the DMM prediction mode is further optimized.