On segmentation with Markovian models

Abstract

This paper addresses the image modeling problem under the assumption that images can be represented by 2d order, hidden Markov random fields models. The modeling applications we have in mind com- prise pixelwise segmentation of gray-level images coming from the field of Oral Radiographic Differential Diagnosis. Segmentation is achieved by approximations to the solution of the maximum a posteriori equation (MAP) when the emission distribution is assumed the same in all models and the difference lays in the Neighborhood Markovian hypothesis made over the labeling random field. For two algorithms, 2d path-constrained Viterbi training and Potts-ICM we investigate goodness of fit by study- ing statistical complexity, computational gain, extent of automation, and rate of classification measured with kappa statistic. All code written is provided in a Matlab toolbox available for download from our website, following the Reproducible Research Paradigm.