Automatic Machine Learning of Decision Rules for Classification Problems in Image Analysis

A new method for automatic machine learning of decision rules for classification problems in image analysis is presented. The method aims at simultaneous decision rule inference and selection of discriminative features which characterize the image entities to be classified. The method is based on the approximation of class conditional densities by a mixture of parametrized densities of a special type using the EM algorithm. Its performance is tested on a classification problem involving real image data.