Pixel labelling for three-dimensional scenes based on Markov mesh models

Abstract The paper discusses the problem of pixel labelling, for three-dimensional images. It is assumed that contextual information in the picture can be modelled by a third-order Markov mesh. The model and associated methodology are generalisations, from two to three dimensions, of work of Lacroix. The algorithm developed is therefore a generalisation of Devijver's F-G-H algorithm. The paper sets out the labelling algorithm and repoets the results of simulation experiments to explore its performance in practice.