A New Virtual View Rendering Method Based on Depth Image

Virtual view rendering has been considered as a key technique in three-dimensional display. A new virtual view rendering method aimed at flat-panel auto-stereoscopic displays which shifts each pixel in reference image horizontally based on depth image is proposed in this paper. Firstly, the depth image is pre-processed by expanding objects boundaries so as to avoid foreground being split due to the inaccuracy of depth image. The second is to horizontally shift all points in reference image to the desired virtual position based on associated disparity, which is determined by depth image and the limit of screen parallax. Finally, to handle the tradeoff between image quality and algorithm complexity, we fill the small holes with a modified interpolation method while the remaining large disoccluded regions are recovered by a depth-aided inpainting algorithm. Experiments are carried out to show the superiority of the proposed method.