Unsupervised statistical segmentation of multispectral SAR images using generalized mixture estimation

This work deals with the estimation of generalized mixtures with applications to unsupervised statistical multisensor image segmentation. A mixture is said to be 'generalized' when the exact nature of the noise components is not known; one assumes, however, that each belongs to a finite known set of families of distributions. We propose some methods of estimation of such mixtures based on Expectation-Maximization (EM), and Iterative Conditional Estimation (ICE, [6]) algorithms. The set of families of distributions is assumed to lie in Pearson's system.