Decision rules for choice of neighbors in random field models of images

Random field models have many applications in image processing and analysis. The mainconcern of this paper is to design a decision rule for fitting an appropriate random field model to a given image. We assume that the given image is a particular realization of a homogenous Gaussian discrete random field. We represent the underlying random field by a set of parametric models representing the spatial dependence. Using spectral representations of the random field and standard Bayesian methods, we develop a decision rule for choosing an appropriate model from a class of such models. We discuss the relevance of the theory developed in this paper for applications in image modeling and texture characterization.