Abstract
We present a novel method for color image segmentation based on an unsupervised learning model and feature selection. Our focus here is to develop an expectation maximization algorithm based on a mixture of bounded generalized Gaussian model combined with a feature selection mechanism. The developed statistical model offers more flexibility in data modeling than the Gaussian distribution and the feature selection mechanism aims at eliminating irrelevant features and then improving the segmentation performances. Obtained results performed on a large dataset of real world color images confirm the effectiveness of the proposed approach.