Representing probability distributions of image segments and segmentations

The authors develop a method for probabilistically maintaining sets of alternative homogeneous regions and segmentations. Depending on the image size and complexity and on the applications, a probability distribution can be constructed over the entire image, or a distribution over partial segmentations can be formed. The authors develop an efficient representation structure and a probabilistic mechanism for applying Bayesian model-based evidence to guide the construction of the representation and influence the resulting posterior distribution over the space of alternatives. The formalism is applied to range images using a piecewise-planar model with additive Gaussian noise.<
>