Color image segmentation using Gaussian mixtures and particle swarm optimization

Abstract

The model of Gaussian Mixture is particularly useful to perform unsupervised learning. Currently, the principal technique to estimate the mixture parameters is the Expectation Maximization method which has a great chance of obtaining sub-optimal results. In this work we opted, instead, for the Particle Swarm Optimization as an alternative way to estimate parameter of Gaussian Mixture applied to multivariate data, which has greater chance of reaching the optimum. To evaluate the proposed approach, color images from fluorescence microscopy are segmented considering the 3D color space. Some particular features of this kind of color image are also considered to improve the performance of the search. © 2012 Springer-Verlag.