Fast stereo matching via graph cuts

Abstract

In this paper, a novel approach that addresses the stereo correspondence problem is proposed. Some recent techniques that transform the stereo correspondence problem to a minimization of a global energy function are based on graph cuts. These techniques based on graph cuts consider all possible disparities between minimum and maximum values for each pixel. The originality of this work is to construct a reduced graph where only some potential values in the disparity range are selected for each pixel. These values can be found using a local analysis method for stereo matching. This method allows us to reduce the size of the graph, thus reducing computation time and improving the quality of disparity map, and at the same time to make wider the disparity range.