Depth-based View Synthesis Technology In Free-view Video System

With the rapid development of high definition video, traditional2D videosystem can not provide the realism of objects to people, so digital video can reflectnatural things clearly and truly. At the same time,3D can provide realism to people,and free-view system of interactivity can watch advanced visual media at anyposition. What’s more, the system has a good prospect in many aspects, such asremote monitoring, telemedicine, virtual reality, and it is a hot issue at domestic andforeign. Virtual view rendering and related technologies is researched in the paper.Firstly, according to the correlation between time and spatial domain of depthimage, the problem of light-based filtering algorithm in time domain is discussed,and the depth filtering method about edge enhancement is proposed. The edge ofdepth map is focus on protecting edge regions in proposed method, so the subjectivequality of virtual view image can be improved to some extent, and coding efficiencyof depth sequence is enhanced.Secondly, through analyzing the computational complexity of the process invirtual view rendering, a rendering method is proposed by improving view mapefficiency using compression information of coding stream. The coding models ofdepth and color sequences are considered in the method. At first, the renderingimage is divided to motion regions, static regions and edge regions. Next, differentregions are rendered in different precision, which can reduce time of coordinatemapping. Finally, the results are deal with necessary post-processing after rendering,which can solve edge noise problem due to inaccuracy depth information.Experiments show that it can be save rendering time about35%to50%.Finally, the occlusion and exposure problem is primary cause of hole inrendering image which often happens in sudden change regions of depth image and is the junction of background and prospects, layered virtual view rendering schemeis proposed. It can reduce hole of occlusion regions and can solve the problem ofbackground covering prospects in rendering process. At first, the global backgroundof test sequence is extracted. Then, object information of each frame is gained by aseries of processing. Next, different priorities are adopted in rendering process.Experimental results show that the proposed method can effectively reduce the holeof virtual view and solve well the noise problem on background covering prospects.