The generalized fuzzy integral in computer vision

To emulate human reasoning intelligent systems must have the ability to integrate information provided by multiple sources. This information can be used to increase object classification confidence, remove ambiguity inherent in a single representation, and resolve conflict in separate decisions.; In this dissertation, a method of evidence fusion based on the generalized fuzzy integral with respect to a t-conorm based fuzzy measure is developed. The objective evidence, in the form of a fuzzy membership function, is combined, non-linearly, with evaluations of the worth of the sources with respect to the decision. Various theoretical properties of this new technique are developed and a method for calculating appropriate fuzzy density values (the degrees of importance of the sources) from the training data is presented. The applicability of this technique to information fusion in computer vision is demonstrated through simulation and with object recognition data from forward looking infrared and TV multi-sensor imagery.