Bayesian Stereo Matching

Li Cheng, Terry Caelli

Abstract: In this paper we explore a Bayesian framework for inferring the disparity map from an image pair. Markov Chain Monte Carlo sampling techniques are employed for learning the hyper-parameters which control two robust statistical functions for modelling the specific image pair; and loopy belief propagation is used for approximate inference of the MAP disparity map. Encouraging results are obtained on a standard set of image pairs.