Research On Key Tenchnology Of High Degree Of Freedom Video Coding And Rendering

Humans can freely change their position and orientation when viewing scenes.To simulate this feature,High Degree-of-Freedom(HDo F)video collects information about a user's position and orientation through sensors,and renders a target viewport in real time.It introduces binocular parallax and interactive motion parallax that provides the user with a more natural and immersive visual experience.The Moving Picture Experts Group(MPEG),a working group of ISO/IEC,has launched relevant standardization activities to promote application and industrialization of HDo F video.However,there are still many technology difficulties in practical applications of HDo F video.On the one hand,HDo F video requires a large number of cameras to capture the scene,hence huge data costs a tremendous amount of transmission and bandwidth.On the other hand,because of the restriction of storage and processing capacity,the pixel rate to be decoded and rendered and video coding standard are both limited on the receiver side.Therefore,the dissertation studies the key tenchnology of HDo F video,and the main contributions are the following:(1)Aiming at the problem in transmitting a tremendous amount of data in HDo F video,we propose an efficient method to do fusion expression of multiviews,which eliminates the repeated pixels among source views.The method adaptively selects several views with high unpredictable pixel rate as basic views,and use them to prune the repeated pixels in the remaining views.For the unpredicted pixels,we crop them to multiple sub-images.Only basic views and sub-images are transmitted.Experimental results show that,compared with transmitting all source views,the method reduces number of transmitted pixels by 70.6% on average.At the same time,the quality of the rendered target view are improved up to 1.4d B.(2)Aiming at the problem in organization of the cropped sub-images at the block level,we propose a method including sub-image merging,adding constraints of sub-image size and position in the packed views.Matching with coding features such as block partitioning and inter/intra prediction,the method increases coding efficiency.Experimental results show that,Y,U and V components can obtain an average performance gain of 13.37%,3.65%,and 4.16%,respectively,compared with HM16.16 + TMIV2.0,which is the reference software of HDo F video.(3)Aiming at the real-time problem of target view rendering in HDo F video,we propose a method to cull sub-image repidly,which prevents sub-images having no overlap with the target view from view rendering.It can reduce the computational burden of target view rendering.The experimental results show that,for sequences with a large field of view,without changing the quality of the rendered target view,this method culls transmitted pixels by 72.15% and reduces the total processing time by 49.38% on average.It improves the practicality of HDo F video.Finally,the dissertation summarizes the specific contents of each research point and looks forward to the future improvement.