Color–texture segmentation using unsupervised graph cuts

This paper proposes a novel approach to color–texture segmentation based on graph cut techniques, which finds an optimal color–texture segmentation of a color textured image by regarding it as a minimum cut problem in a weighted graph. A new texture descriptor based on the texton theory is introduced to efficiently represent texture attributes of the given image. Then, the segmentation is formulated in terms of energy minimization with graph cuts, where color and texton features are modelled with a multivariate finite mixture model with an unknown number of components. Contrary to previous supervised graph cut approaches, our method finds minimum cuts using split moves in an unsupervised way. The segmentation result, including the number of segments, is determined during the split moves without user interaction. Thus, our method is called unsupervised graph cuts. Experimental results of color–texture segmentation using various images including the MIT VisTex datasets and the Berkeley datasets are presented and analyzed in terms of precision and recall to verify its effectiveness.