An Efficient Spatially Constrained EM Algo - rithm for Image Segmentation

We present a novel EM algorithm for model-based image segmentation which incorporates efficient and economical spatial constrains among pixels via a Markov random field model. We adopt a generative model in which the unobserved class labels of neighboring pixels in the image are assumed to be generated by prior distributions with similar parameters. We derive a penalized log-likelihood optimization procedure for estimating the parameters of the pixels' labels priors, and those of a Gaussian observation model that is shared among pixels. Our algorithm is very easy to implement and is similar to the standard EM algorithm for Gaussian mixtures, with the main difference that the labels posteriors are 'smoothed' over pixels between each E-and M-step by a standard image filter. Experiments on synthetic and real images show that our algorithm achieves competitive segmentation results compared to other Markov-based methods, and is in general faster.