Abstract
An algorithm for fusing multiple handwritten-numeral classifiers is addressed using the fuzzy integral in the sense of adaptive aggregation. This method includes a procedure for calculating the /spl lambda/-fuzzy measures which are adaptively adjusted depending on the interactions among individual classifiers. Based on these fuzzy measures, the fuzzy integral is then used as a nonlinear functional to search for the maximum degree of agreement between the complementary/conflicting multiple sources of evidence. Results obtained from the fuzzy integral are used for decision making in the classification problem. Experimental results on handwritten numeral recognition show that the performance of this multi-classifier fusion method outperforms that of other conventional classifier-combination techniques.