Multiresolution Contourlet Transform Fusion Based Depth Map Super Resolution

With the rapid advancement of 3D applications, depth map super resolution has been a serious problem to be solved. Many researches has proposed depth map super resolution methods focusing on spatial domain which can't produce clear edges in the super resolution results. However, different from color images, depth maps have clear edges along them with internal smoothness. In this paper, we propose a novel multiresolution contourlet transform fusion based depth map super resolution method to enhance the quality of depth maps with preserving more contour information. We first transform the depth maps of multiple views via contourlet transform (CT) into multiresolution coefficients. Then we fuse the coefficients of same resolution to get the fused coefficients. Finally the target depth map is upscaled utilizing the prevalent upscaling framework JBU or WMF. Depth map fusion in transform domain is first proposed to improve the quality of the target depth map. A CT based depth map fusion model can not only produce sharper edges, but alleviate the noise in the depth map. Experimental results demonstrate that our proposed method outperforms many state-of-the-art algorithms in both objectively and subjectively.