Greedy Gaussian mixture learning for texture segmentation

Abstract : The problem of segmenting an image into several modalities representing different textures can be modeled using Gaussian mixtures. Fitting a Gaussian mixtures on the data is not trivial problem and no guaranteed optimal algorithm exists. In this paper we show the benefits of a recently developed greedy procedure to Gaussian mixture learning to the problem of texture segmentation. We present the greedy learning method and provide experimental results illustrating the potential of the new method.