Intensity and feature based stereo matching by disparity parameterization

In this paper, we propose a new solution to the stereo correspondence problem by including features an intensity based matching. The features we use are intensity gradients in both the x and y directions of the left and the deformed right images. Although a uniform smoothness constraint is still used, it is nevertheless applied only to non-feature regions. To avoid local minima in function minimization, we propose to parameterize the disparity function by hierarchical Gaussians. A simple stochastic gradient method is used to estimate the Gaussian weights. Experiments with various real stereo images show robust performances.