Priors and constraints in Bayesian image segmentation based on finite mixtures

In this paper, we propose the use of prior densities within the framework of finite mixture models applied towards image segmentation. We pose segmentation as a pixel labeling problem and investigate a generalized expectation maximization algorithm for the Bayesian estimation of the pixel labels. This algorithm is based on a unique spatially-variant mixture model and has the flexibility of incorporating any useful prior information on the potential label configurations. Specifically, two different priors are proposed for pixel labeling and their effectiveness is assessed quantitatively on simulated images at various noise levels. A qualitative evaluation has also been performed using clinical magnetic resonance images of the human brain.