Texture segmentation using Kohonen self organizing feature map

This paper presents a range of techniques for texture segmentation using Kohonen self organizing feature maps (KSOFM). Various features, based on fractals and energy measures, are extracted from the texture images and then applied to the neural network. First, local features are extracted from the whole image using overlapping windows, then the feature vectors of each window are placed into the feature space. KSOFM is used as a clustering technique to assign a label to the groups that have similar features. A smoothing algorithm is performed followed by edge detection to give an edge map of the segmented image. An integrated software has been developed to illustrate the complete process of texture segmentation.