Image Processing Techniques for Unsupervised Pattern Classification

Abstract

All the clustering methods presented in this chapter tend to generalize bi-dimensional procedures initially developed for image processing purpose. Among them, thresholding, edge detection, probabilistic relaxation, mathematical morphology, texture analysis, and Markov field models appear to be valuable tools with a wide range of applications in the field of unsupervised pattern classification. Following the same idea of adapting image processing techniques to cluster analysis, one of our other objectives is to model spatial relationships between pixels by means of other textural parameters derived from autoregressive models (Comer & Delp, 1999), Markov random fields models (Cross & Jain, 1983), Gabor filters (Jain & Farrokhnia, 1991), wavelet coefficients (Porter & Canagarajah, 1996) and fractal geometry (Keller & Crownover, 1989). We have also already started to work on the adaptation of fuzzy morphological operators to cluster analysis, by extracting the observations located in the modal regions performing an