Output details

A `no panacea theorem' for classifier combination

Significance of output:

<24>The limits of classifier combinations are specified by a theorem which is stated and proved under specific assumptions. A series of rules for classifier combination were developed in earlier work by Josef Kittler, who provided empirical support for their applicability in various conditions. The work in this paper gives formal results with which insights are obtained on the performances of such fusion rules. With these results, by checking the statistical assumptions that underlie the environment, we are able to make statements about how well classifier fusion might perform (without empirical evaluations).