The EM/MPM algorithm for segmentation of textured images: Analysis and further experimental results

In this paper, we present new results relative to the "expectation-maximiza tion/maximization of the posterior marginals" (EM/MPM) algorithm for simult aneous parameter estimation and segmentation of textured images. The EM/MPM algorithm uses a Markov random field model for the pixel class labels and alternately approximates the MPM estimate of the pixel class labels and est imates parameters of the observed image model. The goal of the EM/MPM algor ithm is to minimize the expected value of the number of misclassified pixel s. We present new theoretical results in this paper which show that the alg orithm can be expected to achieve this goal, to the extent that the EM esti mates of the model parameters are close to the true values of the model par ameters. We also present new experimental results demonstrating the perform ance of the EM/MPM algorithm.