ADVANCES IN FUZZY INTEGRATION FOR PATTERN-RECOGNITION

Uncertainty abounds in pattern recognition problems. Therefore, management of uncertainty is an important problem in the development of automated systems for the detection, recognition, and interpretation of objects from their feature measurements. Fuzzy set theory offers numerous methodologies for the modeling and management of uncertainty. One such fuzzy set theoretic technology which has proven quite useful in pattern recognition is the fuzzy integral. The purpose of this paper is to examine new utilizations of the fuzzy integral as a decision making model in the area of object recognition. In particular, we develop generalizations of the fuzzy integral and show that these generalizations can achieve higher recognition rates in an automatic target recognition problem. Also, we demonstrate significant increases in recognition rates using the fuzzy integral to fuse the results of different neural network classifiers in a complex handwritten character recognition domain.