Research On Depth-image-based Rendering And Distortion Model

Free viewpoint video represents a new technology for representation of moving 3D scene content.It allows the user to watch the data from any viewpoint and thus provides direct interaction with the scene instead of passive video consumption.Due to the development of 3D display and interactive multi-media systems,new 3D video applications,such as 3DTV,remote video conference,3D video surveillance,and free viewpoint video,are attracting significant interests.In this paper,virtual view synthesis and its distortion analysis as current research hot spot,have attracted increasing researchers inside and outside country.The main work as follows:We proposed a novel DIBR method.First,a luminance compensation algorithm based on histogram matching is applied on the two reference views to reduce the illumination difference.Second,an isolated depth pixel removing method is proposed to refine the depth map.Third,the 3D warping with depth-based pixel interpolation method is adopted to eliminate the small cracks in the virtual view image.Finally,proposed background-based hole filling method.Distortions in depth map may introduce pixels position errors,and the pixels position errors will lead to geometry distortion of the synthesized views.In this paper,we explore the maximum tolerant depth distortion(MTDD)which can retain the quality of synthesis view without any changes.Experimental results show that depth distortion within the range of MTDD the quality of virtual view images without any texture and geometry distortions.We divide depth map into sharped boundary regions and smoothed non-boundary areas with the method that boundary segmentation based on the just noticed depth difference(JNDD),and assign the quantization parameter for the segmented regions,respectively.The experimental results show that the proposed method can retain the edge information,and effectively reduce the distortion of rendering virtual view.