Kolmogorov and Zabih's Graph Cuts Stereo Matching Algorithm

Binocular stereovision estimates the three-dimensional shape of a scene from two photographs taken from different points of view. In rectified epipolar geometry, this is equivalent to a matching problem. This article describes a method proposed by Kolmogorov and Zabih in 2001, which puts forward an energy-based formulation. The aim is to minimize a four-term-energy. This energy is not convex and cannot be minimized except among a class of perturbations called expansion moves, in which case an exact minimization can be done with graph cuts techniques. One noteworthy feature of this method is that it handles occlusion: The algorithm detects points that cannot be matched with any point in the other image. In this method displacements are pixel accurate (no subpixel refinement). Source Code The software rewritten from Kolmogorov’s code is available at the IPOL web page of this article1. A set of stereo pairs is available and Kolmogorov and Zabih’s algorithm can be tried on line. In the demo, the algorithm is run on six overlapping slices of the images, for efficiency purpose. Essentially, two parameters are needed: K associated to occlusion cost and λ to data fidelity. By default they are tuned automatically but they can be adapted to get better results. Supplementary Material In the demo, an optional rectification step can be launched before running the algorithm. The source code for this preprocessing step (not reviewed) can be found at the IPOL web page of this article2.