Numerical Pattern Classification

Abstract

Numerical pattern classification deals with the problem of assigning feature vectors c to a class Ω _K of the available classes Ω = {Ω₁, Ω₂,..., Ω _K }. The features are computed from (noisy) sensor data, like images or speech signals. We postulate that signals can be associated with features which allow the classification, i.e., features of different classes should be different and separated from each other. Features belonging to the same class are thus expected to occupy a compact area of the feature space [Nie90a].