An efficient spatially constrained EM algorithm for image segmentation
Aristeidis Diplaros, Th. Gevers, Nikos Vlassis · UvA-DARE (University of Amsterdam) · 2005
We present a novel EM algorithm for model-based image segmentation which incorpo-rates efficient and economical spatial constrains among pixels via a Markov random fieldmodel. We adopt a generative model in which the unobserved class labels of neighboringpixels in the image are assumed to be generated by prior distributions with similar para-meters. We derive a penalized log-likelihood optimization procedure for estimating theparameters of the pixels’ labels priors, and those of a Gaussian observation model thatis shared among pixels. Our algorithm is very easy to implement and is similar to thestandard EM algorithm for Gaussian mixtures, with the main difference that the labelsposteriors are ‘smoothed’ over pixels between each E- and M-step by a standard imagefilter. Experiments on synthetic and real images show that our algorithm achieves com-petitive segmentation results compared to other Markov-based methods, and is in generalfaster.Keywords: Image segmentation, Hidden Markov random fields, EM algorithm, Boundoptimization, Spatial clustering.