A stochastic model for image segmentation involving constrained least squares estimation

Abstract

The aim of the paper is to outline a layered statistical image model suitable for unsupervised image segmentation. The segment internal texture signal is described based on its spatial frequency representation while the image partition is modelled as a sample of a Gibbs/Markov random field. The most likely segmentation is estimated using a maximum a posteriori (MAP) formulation with the unknown parameters being determined by constrained least squares (CLS) estimation.<>