Depth map refinement using superpixel label information
Su-Min Hong, Yo‐Sung Ho · 2016
In this paper, we propose a depth map refinement method to reduce the mismatch depth values along object boundaries. We design a filter which has segmentation, distance and color similarity weighting function. Our segmentation weighting term is defined based on SLIC superpixels because it delivers sufficient results at a very high speed. We give a penalty factor for the neighborhood pixels that are not within the same superpixel. Experimental results show the proposed method solve the problem that depth value propagation from one region to another region. So, the proposed method efficiently enhance the depth map quality. Experimental results shows the proposed method outperforms the conventional algorithms in terms of the RMSE for all the tested images. Therefore, they are expected to be used for various applications of 3D video processing.