An Unsupervised, Agglomerative, Spatially Aware Texture Segmentation Technique

A novel method of performing multiscale, hierarchical segmentation of images using texture properties is introduced. The images are first requantized using contiguity-enhanced K-Means clustering. Morphological operations and region growing based on textural properties are used to arrive at the most detailed segmentation. Successively coarser segmentations are achieved by the use of inter-cluster distances in a dyadic, agglomerative technique. An objective way of quantitatively measuring the performance of texture segmentation algorithms independent of texture classification or training is also introduced. The method described in this paper is compared with some unsupervised texture segmentation algorithms reported in the literature. Our method performs better than other unsupervised and untrained texture segmentation algorithms on certain kinds of textured images. Results are presented on natural textures and on real-world scenes.