Stereoscopic image generation based on weighted iteration of depth image

In Depth-Image-Based-Rendering (DIBR) system, right images (left images) can be rendered by 3D-Warp. However, a large problem which is so-called disocclusion will exists in generated images. The previous work uses Gaussian filter to smooth depth maps to reduce the number of big holes, and then adopts an average filter to fill existing holes. However, lots of references show that this method will lead to geometric distortion at vertical edges in rendered images. In order to solve this problem, our paper proposes a weighted iteration method to blur the depth map, and then adopts the expansion method to fill holes. Experimental results demonstrate that the proposed method gets high quality in warped image and the PSNRis better than the previous.