Virtual View Generation Method without Pre-processing Depth-Image for 3DTV

Depth image based rendering (DIBR) algorithm can theoretically render arbitrary view, but with low quality and large disocclusion regions. The most common method is that:depth-image is firstly pre-processed and finally the disocclusion regions are inpainted. Preprocessing depth-image leads to serious deformation of virtual view image. Algorithms for image inpainting are sophisticated and fail to obtain ideally results. In this paper, a double-side DIBR method without preprocessing depth-image is proposed. An algorithm for reducing tiny holes is utilized during the process of image rendering,and algorithm for image merging is improved. The experimental results show that deformation is fundamentally eliminated and the complication for inpainting disocclusion regions is also greatly decreased.