Virtual View Synthesis In Freeview Television System

Freeviewpointtelevision(FTV)system,asthebreakthroughofthree-dimensionaltelevision (3DTV) system, canprovide themost naturalviewingexperience. Bysimul-taneouslydisplayingmulti-view(N≥9)videos, itcannotonlymakemultipleviewersto enjoy the TV contents, but also generate the continuous parallax feld in small re-gions to support the head motion of viewers. In FTV system, because of the limitationof bandwidth and storage, only two or three views' videos and depth information aretransmitted and the rest of views are mainly synthesized by depth image based render-ing, DIBR.DIBR has been the key technology in the synthesis of the virtual view images inFTV system. Despite of its advantages, DIBR also has some drawbacks. It is the errorsof the depth maps and the changing of occlusion relations between objects in diferentviewpointsthatwillcauseholes, ghostcontoursandgeometricdistortionsinthevirtualview images. By focusing on the rendering of the virtual view images in FTV system,this paper proposes the method of virtual view synthesis based on depth correction andDIBR.In this paper, the virtual view images synthesis system can be divided into twoparts, thetransmittingandreceivingend. Inthetransmittingend, basedonthelowrankapproximation model and Mean Shift algorithm, we separate the foreground movingobjectsandthebackgroundandcorrecttheforegroundlosserrorandbackgrounddeptherror in the depth maps. The depth correction algorithm can overcome the problems ofgeometric distortions and image gaps in the virtual view images due to the errors of thedepth maps. Meanwhile, we apply the color correction of the reference images fromdiferentviewpointstoalleviatetheimpactofthechromaticitydiferenceinbackgrounddue to the image blending.In the receiving end, we separate the reference images into three layers, including non-edge, foreground edge and background edge. We warp the layered image withdiferent sampling rates for edge and non-edge layers. When we blend the multiviewimages, the non-edge layer is frstly involved, then the foreground edge and the last isthe background edge. This blending method will reduce the ghost contours and holesaround the objects' boundaries. According to the depth information of backgroundobtainedfromthelowrankapproximationmodel, weproposetheholefllingalgorithmbased on back-warping.Experimental shows that the method of virtual view synthesis based on depth cor-rectionandDIBRsolvestheproblemofholes,ghostcontoursandgeometricdistortionsand can be applied in the FTV system to synthesize high-quality virtual view images.