Sharp disparity reconstruction using sparse disparity measurement and color information

Recently, the work on dense disparity map reconstruction from 5% sparse initial estimates containing edges in disparity, has been proposed [1]. Practically, however, edges in disparity is unknown unless a dense disparity map has already been generated. In this paper, we present a realistic reconstruction framework for obtaining sharp and dense disparity maps from fixed number of sparse initial estimates with the aid of color image information. Experimental results show that sharp and dense disparity maps can be reconstructed at the cost of one pixel accuracy.