A new paradigm for the description of image patterns-from pixels to fuzzy sets of rules

A paradigm for the description of image patterns is presented: it is proposed that images are described by fuzzy IF...THEN rules instead of pixel values. This approach may benefit from recognised fuzzy systems' superior incorporation of measurement uncertainties, greater resources for managing complexity and better ability to deal with natural language. The concept of relevance has been proposed as a measure of the relative importance of sets of rules (Salgado, 1999, and Salgado et al., 2000). Based on this concept a new methodology was developed: SLIM (separation of linguistic information methodology) (Salgado, 1999, and Salgado et al., 2000). An algorithm implementing SLIM is presented in this paper, derived from the fuzzy C-means clustering algorithm, here applied to organise the fuzzy IF...THEN rules that describe the image. The Lena and the Abington Cross images have been successfully used to illustrate the identification process. The proposed SLIM algorithm has been successfully applied to illustrate a segmentation operation in the "fuzzy rules domain", using the Abington Cross image.