Unsupervised texture segmentation based on immune genetic algorithms and fuzzy clustering

We consider a new, adaptive approach to unsupervised textured region segmentation. There are three phases within each iteration of the process: (1) Gabor filter based feature extraction; (2) Fuzzy clustering of texture homogeneity to yield a spatial segmentation; and (3) An optimization procedure to update the filter parameters. The selection objective used for filter optimization was calculated using the maxmin principle on the output from the Fisher function. This enabled the energy distributions of the distinctly textured sub images to be well separated. Experimental results demonstrated the effectiveness of the proposed approach.