Generation of fuzzy rules involving spatial relations for computer vision

Rule-based systems are commonly used in computer vision for scene analysis. In this paper, we propose a method for generating fuzzy IF-THEN type rules involving spatial relationships between labeled regions automatically from training data. The proposed method consists of five stages: I) determination of membership functions for representation of spatial relations, II) extraction of training data that represent spatial relations, III) estimation of class relation membership functions, IV) elimination of redundant relations, and finally V) generation of rules. Results are shown for two examples