Decision Combination in Multiple Classifier Systems：（决定在多个分类器组合系统）.pdfVIP

66 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 16. NO. I . JANUARY 1994 Decision Combination in Multiple Classifier Systems Tin Kam Ho, Member, IEEE, Jonathan J. Hull, Member, IEEE, and Sargur N. Srihari, Senior Member, IEEE Abstract-A multiple classifier system is a powerful solution to for problems with many classes, as correct rate at top choices difficult pattern recognition problems involving large class sets drops, secondary choices often contain near misses that should and noisy input because it allows simultaneous use of arbitrary not be overlooked. In [33], all possible combinations of top feature descriptors and classification procedures. Decisions by choices for a given set of classes are examined. A test is the classifiers can be represented as rankings of classes so that they are comparable across different types of classifiers and proposed for the dominance of each true class occurring with different instances of a problem. The rankings can be combined each combination that would justify the final assignment of by methods that either reduce or rerank a given set of classes. the input to that class. Reliable decisions are achieved by ex- An intersection method and a union method are proposed for haustive enumeration, and therefore the method is expensive. class set reduction. Three methods based on the highest rank, For n classes and IC classifiers, 7hk combinations need to be the Borda count, and logistic regression are proposed for class set reranking. These methods have been tested in applications on