Free Viewpoint Video Coding And Synthesizing

As video technologies develops,MVD(Multi-View plus Depth)format has been widely applied in FVV(Free Viewpoint Video)system.Compared with the fixed-viewpoint single/multiple view format,MVD format can provide arbitrated'view videos to the user via the DIBR(Depth-Image-Based Rendering)technique and improves the experiences of users.In this paper,FVV system is chosen as our research target.Depth information quantization,view synthesis and view synthesis distortion estimation are as three crucial techniques of the FVV system are carefully studied.The main contributions and creative research results include:· We develop a nonlinear depth quantization algorithm towards view synthesis.The common depth quantization methods are firstly analyzed.Then,the reason of depth quantization caused view synthesis distortion are given out,namely the misalign-ment between warping interval and resolution interval.According to this,an ad-justment is added in the traditional nonlinear depth quantization to align these two intervals.Experimental results have verified that our proposed method obliviously reduces the storage of depth without introducing view synthesis distortion.· We develop a region-aware fast view synthesis algorithm.The low accuracy of region calculation model is firstly analyzed as the main reason that decreases the quality of view synthesis.A much more generalized region calculation model is then proposed,which can accurately calculate the size of regions via an iterative ap-proach.After that,the complicated formulas in iterative approach are transformed and simplified to disparities formulas,which can be easily calculate by introduc-ing a separable lookup-table technique.Experimental results have verified that our developed algorithm achieves significant computation savings without sacrificing synthesis quality.· We develop a depth-bin-based graphical model(DBGM)for fast view synthesis dis-tortion estimation(VSDE)algorithm.The complicated VSDE model is commonly represented at depth-value-level,which is firstly analyzed as the main reason that limits its efficiency.Then,depth bin is defined,and its properties are carefully stud-ied.After that,the VSDE model is represented and optimized at depth-bin-level.Experimental results verified that depth-bin-based graphical model has much more efficiency for the VSDE.· We develop a pixel-level view synthesis distortion estimation algorithm for region-based 3-D video coding.Pixel-level view synthesis distortion is first defined.To estimate it,a backward prediction method is then developed,which starts from the pixels of interest(POIs)in the virtual view and finds their corresponding pixels in the reference view via a coarse-to-fine approach.Experimental results has veri-fied that our proposed algorithm accurately achieves fast pixel-level view synthesis distortion estimation.