Unsupervised texture image segmentation

Abstract

This paper is focused on a hierarchical structure of modular self-organizing neural networks for unsupervised texture segmentation (sofm-nn). Input data consists of local information regarding textures (cooccurrence matrix elements) and the texture image itself. An unsupervised segmentation is done using a sofm-nn network and then the final segmentation is performed by another sofm-nn neural network using the previously obtained results. Experimental results show the efficiency of the proposed method.