High-precision stereo disparity estimation using HMMF models

In this paper, stereo disparity reconstruction is formulated as a parametric segmentation problem in a Bayesian framework: the goal is to partition the reference image into a set of non-overlapping regions, inside each one of which a specific disparity model (which consists of two coupled membranes) is adjusted. The problem of simultaneously finding the regions and the parameters of the corresponding models is formulated using a novel probabilistic framework which uses a hidden Markov random measure field model, which allows one to efficiently find the optimal estimators by minimization of a differentiable cost function. This framework also allows for the explicit modeling of occlusions, consistency constraints and correspondence of disparity and intensity discontinuities. It is shown experimentally that this method produces competitive results, with respect to state-of-the-art methods, for discretized (integer) disparities and significantly better results for high-precision real-valued disparities. © 2006 Elsevier B.V. All rights reserved.

Bayesian method; Disparity map; Doubly stochastic prior model; Image segmentation; Markov random fields; Occluded regions; Parametric disparity model; Stereo correspondence; Subpixel disparity values Image segmentation; Markov processes; Mathematical models; Probability; Problem solving; Stereo vision; Bayesian method; Doubly stochastic prior model; Markov random fields; Parametric disparity model; Subpixel disparity values; Image reconstruction

VIVO