Stereo Reconstruction Using High Order Likelihood

Abstract

Under the popular Bayesian approach, a stereo problem can be formulated by defining likelihood and prior. Likelihoods are often associated with unary terms and priors are define by pair-wise or higher cliques in Markov random field (MRF). In this paper, likelihood is proposed using higher order cliques. Numerous patch based matching methods such as normalized cross correlation, Laplacian of Gaussian, or census filters are under the naive assumption that a patch’s pixels all have same disparities. However, patch-wise cost can be formulated as higher order clique for MRF so that the matching cost is a function of image patch’s disparities. A patch obtained from a projected image by disparity map should provide a better match without the blurring effect around disparity discontinuities. Among patch-wise matching costs, census filter approach can be easily reduced to pair-wise cliques. The experimental results on census filter high older likelihood demonstrate the advantages of high order likelihood over independent identically distributed unary model. 1.