We address the problem of unsupervised segmentation of textural images by relying on morphological image representations and active contours. More precisely, starting with the assumption that textures are statistical ensembles of local image structures, known as textons, we first suggest to represent a texture image through a tree of ellipses, which are derived from the level lines of the image; we then investigate the statistical properties on the tree of ellipses and finally achieve a

more »

... tion of the texture image by grouping all the ellipses into several subsets according to some statistical measurements. The grouping process is formulated as an energy minimizing problem and the solution is obtained by evolving an active contour based on Kullback-Leibler (KL) divergence through a fast global minimization method. Thanks to the proposed ellipsebased features, the segmentation method can integrate local and global information in the image. The experiments on both synthesized and natural texture images validate the approach.