DECISION COMBINATION IN MULTIPLE CLASSIFIER SYSTEMS

A multiple classifier system is a powerful solution to difficult patte rn recognition problems involving large class sets and noisy input bec ause it allows simultaneous use of arbitrary feature descriptors and c lassification procedures. Decisions by the classifiers can be represen ted as rankings of classes so that they are comparable across differen t types of classifiers and different instances of a problem. The ranki ngs can be combined by methods that either reduce or rerank a given se t of classes. An intersection method and a union method are proposed f or class set reduction. Three methods based on the highest rank, the B orda count, and logistic regression are proposed for class set reranki ng. These methods have been tested in applications on degraded machine -printed characters and words from large lexicons, resulting in substa ntial improvement in overall correctness.