A novel approach for depth image based rendering, based on non-linear transformation of depth values

Depth-image-based rendering (DIBR) is the key technology in a 3DTV system, which generates two or more virtual view images from 2D color image and its associated per-pixel depth image. The most troubling problem in DIBR is the appearance of holes in the virtual view images, which are called “disocclusion”. To remove the newly exposed regions, in this paper, we propose a novel approach to synthesize the virtual view images. First, the depth image is pre-processed by an adaptive edge-dependent asymmetric smoothing filter. Secondly, the horizontal displacement of pixels, as the result of 3D image warping, is obtained based on the non-linear adaptive transformation of depth values to depth distances. The experimental results show that the proposed method can efficiently reduce or completely remove the disocclusion and no geometric distortion arises.