Research On The Optimization Method Of Intra Coding In 3D-HEVC

With the maturity of digital video technology and the rapid development of image graphics technology and computer vision technology,3D video has attracted wide attention for its real visual experience,but its huge amount of data brings great pressure to the transmission and storage of video.In order to achieve efficient compression of 3D Video and guarantee the Video quality,the joint collaborative team on video oding proposed a coding standard for 3D video based on traditional 2D video coding standard HEVC,names 3D High Efficiency Video Coding.3D-HEVC adopts the encoding format of multi-view and depth map,and introduces a variety of new technologies on the basis of HEVC to improve the efficient compression of 3D video.However,there is still a problem of high computational complexity in intra coding,which hinders the popularization of 3D video.Therefore,this paper focuses on the optimization method of intra coding in 3D-HEVC to reduce the computational complexity of intra coding,which is of great significance to the popularization of 3D video.Based on the in-depth study of the framework,basic knowledge and coding technology of HEVC and 3D-HEVC,this paper optimizes and improves the intra coding method in 3D-HEVC.The main research contents and innovation points are summarized as follows:Aiming at the problem of high computational complexity caused by recursive quadtree division of depth map during intra prediction,this paper proposes a fast intra depth selection algorithm based on deep learning.Firstly,a Fast Selecting Cu's Depth-Convolutional Neural Network model suitable for optimal depth prediction of3D-HEVC depth maps is built,and then the network model is combined with LCU recursive partition to predict the optimal partition depth of LCU,Finally,according to the predicted results,the top-down quadtree partitioning process and DMM Mode exploration process are terminated in advance.Compared with the 3D-HEVC coding platform,the algorithm in this paper reduces the coding time by 42.58% on average,and the average BDBR is reduced by 0.036%.Meanwhile,compared with the current several popular algorithms,the time saving is improved by 15% on average,and the intra coding in 3D-HEVC is optimized and improved while the quality is guaranteed.In order to solve the problem of high computational complexity caused by the recursive partition of the whole LCU quadtree in intra prediction,this paper proposes an intra fast selection algorithm using decision tree.Firstly,a machine learning model for optimal depth prediction of 3D-HEVC texture map was built,then the model is combined with the quadtree partition process of texture map LCU,and the optimal partition depth of current texture map LCU is predicted based on features such as spatial correlation and texture complexity of texture map coding.According to the predicted results,the LCU partition process of texture map is terminated in advance,finally,based on the correlation between the depth map and the texture map and the characteristics of the depth map,the recursive partitioning process of the depth map LCU and the DMM mode exploration process are terminated in advance.Experimental results show that compared with the 3D-HEVC coding platform,the algorithm in this paper saves 37.05% of coding time on average,while the BDBR only increases by 0.41%.Compared with the current popular algorithm,the average coding time saved is more than 7.35%,effectively realizing the optimization of intra coding in 3D-HEVC while ensuring the quality.