Region-based Stereo Vision using a Minimization Approach
Angeles López, Filiberto Pla

In this paper, a region-based method for stereo vision is presented. The method computes depth directly, without the intermediate calculation of disparities, by the minimization of a correlation-based energy function. Some constraint on the distribution of depth in the regions is assumed, so that the obtained solution is smooth inside the regions while depth discontinuities are allowed in their boundaries. The derived equation indicates whether an increment or decrement in depth is needed in an iterative process in order to achieve the solution. A multiscale approach is used in order to avoid local minima.

Keywords: region-based stereo vision, energy minimization, correlation similarity, depth map.