Image segmentation using simple Markov field models

Abstract

By modelling a picture as a two-state Markov field, MAP estimation techniques are used to develop suboptimal but computationally tractable binary segmentation algorithms. The algorithms are shown to perform well at low signal-to-noise ratios, and analytical procedures are developed for estimating the Markov field transition probabilities. In addition, extensions of this approach to the multispectral and multiregion cases are discussed.