We develop a common theoretical framework for combining classifiers which use distinct pattern representations and show that many existing schemes can be considered as special cases of compound classification where all the pattern representations are used jointly to make a decision. An experimental comparison of various classifier combination schemes demonstrates that the combination rule developed under the most restrictive assumptions-the sum rule-outperforms other classifier combinations schemes. A sensitivity analysis of the various schemes to estimation errors is carried out to show that this finding can be justified theoretically.

We develop a common theoretical framework for combining classifiers which use distinct pattern representations and show that many existing schemes can be considered as special cases of compound classification where all the pattern representations are used jointly to make a decision. An experimental comparison of various classifier combination schemes demonstrates that the combination rule developed under the most restrictive assumptions-the sum rule-and its derivatives consistently outperform other classifier combinations schemes. A sensitivity analysis of the various schemes to estimation errors is carried out to show that this finding can be justified theoretically.

We present in this paper a shape representation that combines both structural and numerical information about each object. Each of these sources of information has its own advantages but in general they are not compatible with each other. In our work we show how it is often possible and beneficial to combine these two types of representation especially for complex objects that have some natural structure. The method is successfully applied to the problem of totally unconstrained handwritten character recognition. Combined with a simple classification scheme this representation yields a good performance both in speed and accuracy.