Comparison Of Several Fusion Paradigms Applied To Pixel-Based Image Classification

This paper is concerned with the development of FuRII, a pixel-based image classification tool developed at DRDC Valcarder. FuRII is based on fuzzy sets and evidence theories and is implemented as an ENVI toolbox The aim with this tool is to compare several fusion operators and rules in the context of image classification applied to land cover mapping. Several fuzzy fusion operators (conjunctive, disjunctive, adaptive and quantified adaptive fusion) and evidential fusion rules (Dempster, Dubois and Prade, Yager and Smets) are tested FuRII permits to model imprecise knowledge with membership functions and fusion can be performed directly with membership values or with mass functions. In this later case, a transformation of membership values into basic belief values is computed Finally, FuPII permits integration of source reliability into the fusion process.