Texture Segmentation by Biologically-Inspired Use of Neural Networks and Mathematical Morphology

The segmentation of image objects by humans has recently been modelled as a two-step process by the BCS/FCS model of Grossberg. First, regions of homogeneous greyvalue distribution or with similar texture patterns are recognized. Second, these regions are progressively grown until they fill-in the whole scene or image. Object boundaries are defined when growing regions with different characteristics meet. In this paper, we use this approach in texture segmentation. We show, that the marker-controlled segmentation based on the watershed transformation is best suited for implementing this model. In order to generate appropriate marker and edge images for a wide variety of input images, we present the texture segregation/region growing approach which extends the conventional feature classification approach of texture analysis.