TEXSOM: Texture segmentation using self-organizing maps

This article describes the so-called TEXSOM-architecture, a texture segmentation architecture based on the joint spatial/spatial-frequency paradigm. In this architecture the oriented filters are automatically generated using the adaptive-subspace self-organizing map (ASSOM) or the supervised ASSOM (SASSOM) neural models. The automatic filter generation overcomes some drawbacks of similar architectures, such as the large size of the filter bank and the necessity of a priori knowledge to determine the filters' parameters. The quality of the segmentation process is improved by applying median filtering and the watershed transformation over the pre-segmented images. The proposed architecture is also suitable to perform defect identification on textured images. (C) 1998 Elsevier Science B.V. All rights reserved.