A variational co-training framework for remote sensing image segmentation

Inspired by the idea of co-training algorithm, in this paper we propose a novel remote sensing image segmentation approach using co-training strategy under variational Bayesian (VB) framework. Image data are characterized in two distinct views, i.e. two disjoint feature sets. A Gaussian mixture model (GMM) is employed for each view. On one hand, underlying structure of image content is inferred automatically with the factor analysis techniques. On the other hand, parameters are estimated in a bootstrap mode with the co-training strategy. In this manner, a satisfying performance can be achieved. Experimental analyses carried out on several different sets of high resolution optical images validate the proposed algorithm.