Un modèle de mélange pour la segmentation de données spatiales

A mixture-model-based approach is introduced in this article for spatial data modeling and segmentation. The spatial dependence of the data is taken into account through the proportions of the mixture, which are modeled as logistic transformations of polynomial functions of the spatial coordinates. The parameters of the proposed model are estimated by maximizing the likelihood criterion through a specific EM algorithm which incorporates a Newton-Raphson algorithm dedicated to the estimation of the logistic functions coefficients. The experiments conducted on synthetic images have shown encouraging results in term of accuracy of segmentation.