Research On Video Coding Optimization And Method

With the development of imaging equipment,people pay more and more attention to visual experience.The data of high-definition video and multi-view plus deep(MVD)stereoscopic video is enormous,and there is a great demand for a more effective compression technology because of the problems of transmission and storage.4K video can bring more excellent definition,more exquisite textures and better performance of fast motion.However,the amount of data that comes with 4K video is huge,and next video coding standard H.266 is already in full swing.The depth map represents the distance from the object to the camera.The multi-view plus depth video which can provide virtual view generation via a Depth Image-Based Rendering technique brings more stereoscopic and more realistic visual experience,and the data of MVD is also huge.The three-dimensional video coding standards AVS2-3D(Audio Video Coding Standard-3D)and 3D-HEVC(3D High Efficiency Video Coding)have been set for the compression of MVD.In this thesis,some techniques in the above-mentioned standards are introduced,and these techniques are improved.The innovations and contributions include the following three points:(1)For the intra Planar prediction mode,according to the shortcomings of inaccurate filling of the bottom row and the right column during horizontal and vertical filtering,a Planar compensation mode is proposed,which add the top left corner of the current coding block to prediction.This method improves the compression performance of sequence with a large area of gently varying regions.(2)This thesis combines video prediction technology and video coding technology and generates one predicted image by optical flow and deep learning with the images that have been coded and reconstructed.The predicted image is called is used as a high performance reference frame.This method integrates the pixels that can be referenced and brings some improvement of coding performance.(3)Compared to natural images,depth maps contain large plain regions surrounded by sharp edges at the object boundaries.Base on Wedgelet Partition,Bi-Parabola Partition Mode(BPPM)is introduced.BPPM can fit the boundary of depth map better.And then we optimize performance and computational complexity.Finally,in the case of low computational complexity,the coding performance of the depth maps is improved,thereby improving the quality of the virtual view synthesis.