A minimum distance feature effectiveness criterion (Corresp.)

A new criterion for measuring separability of two pattern classes is given that is based on the determination of an averaged minimum distance from the mean feature vectors of two classes to an optimal separating hypersurface determined by Baye's postulate.