Unsupervised texture segmentation using tuned filters in Gaborian space

This paper presents a texture segmentation algorithm based on the multi-channel filtering theory. The channels are characterized by a bank of Gabor like tuned modulated basis filters. We have chosen scale-changeable exponential bases of compact support to derive such filters. It is seen that the tuned modulated basis filters closely approximate the Gabor elementary function. Perfect reconstruction of the input image from its filtered images is shown. Computation and storage requirements are considerably reduced. Texture features are obtained by subjecting each (selected) filtered image to a nonlinear transformation and computing a measure of ''energy'' in a window around each pixel. The simple K-means algorithm is used to produce segmentation. (C) 1997 Published by Elsevier Science B.V.