Compact fuzzy rule base generation methods for computer vision

Rule-based approaches are commonly used in vision systems to solve complex problems. The rules are usually elaborated by experts. The authors propose a method to generate such rules automatically from training data. The proposed approach achieves this in four stages: estimation of class membership functions, elimination of redundant features, estimation of the membership functions of linguistic labels that best describe the nonredundant features, and generation of rules. Redundancy detection and rule base generation stages are achieved by training appropriate fuzzy connective-based aggregation networks