Abstract
This is a short tutorial on the concept of the Markov random field (MRF) as applied in computer vision and its relationships with salient terms in the psychology of visual perception, especially connected with Gestalt theory. The concepts discussed are those of graphical structure of an MRF model, variables involved in the model, potentials defined as soft constraints on neighbouring variables, energy and probability functions implied by a network of potentials, and inference procedures aiming at the minimization of energy or, equivalently, the maximization of probability. These concepts are first defined for MRFs in general, then characterized in relation to MRFs associated with vision tasks, then compared with analogous concepts in the psychology of visual perception, and finally evaluated in their flexibility and heuristic power for formal modelling in the psychological study of vision.