Ž . Pattern Recognition Letters 18 1997 1373–1377 Strategies for combining classifiers employing shared and distinct pattern representations 1 J. Kittler ) , A. Hojjatoleslami, T. Windeatt UniÕersity of Surrey, Guildford, Surrey, GU2 5XH, UK Abstract The problem of combining multiple classifiers which employ mixed mode representations consisting of some shared and some distinct features is studied. Two combination strategies are developed and experimentally compared on mammographic data to demonstrate their effectiveness. q 1997 Elsevier Science B.V. Keywords: Classification; Multiple expert function 1. Introduction Combination of classifiers has received consider- able attention in the pattern recognition literature during the last quenquennium. For a number of years various strategies for combining different classifier designs have been used to improve the performance of a pattern recognition system. The advocated clas- sifier combination methodology has been largely heuristic although some attempts to provide a com- mon underlying framework for at least some combi- nation rules have recently been reported. The emerg- ing classification of approaches involves four cate- gories: multiple classifiers using an identical pattern representation; multiple classifiers using distinct pat- tern representations; data dependent combination strategies; multistage classifiers. For the first cate- Ž . gory it has been shown by Tumer and Ghosh 1996 ) Corresponding author. 1 This work has been supported by EPSRC Grant GRrJ89255. for discriminant function classifiers and by Kittler Ž . 1997 for the classifiers approximating a Bayes decision rule by computing the a posteriori class probabilities, that any improvement in performance derives from the well-known principles of the Bayesian estimation theory, i.e. reducing estimation errors by virtue of using a larger number of samples. Ž . For the second category, Kittler et al. 1996 showed that many existing combination schemes can be developed from a common Bayesian framework. This has recently been extended to take into account the confidence of individual experts in the computed a posteriori probabilities. The last two categories of approaches are exemplified by the data-dependent Ž . multiple expert scheme Woods et al., in review , where the decision about the class membership of each unknown pattern is made by the locally most Ž reliable expert, and by the multistage classifiers Ho . et al., 1994; Xu et al., 1992 , respectively. In this paper we shall focus on the first two categories of multiple expert decision making and address the problem of combining classifiers where 0167-8655r97r$17.00 q 1997 Elsevier Science B.V. All rights reserved. Ž . PII S0167-8655 97 00095-0