Adaptation of linear discriminant analysis to second level-pattern recognition classification

Linear Discriminant Analysis (LDA) may be easily adapted to second level-pattern recognition by building hyperspherical boundaries around the class centroid in the discriminant space. Thus, a given test object may belong to one of the studied classes, to more than one class (when situated within overlapped class envelopes) or may be an outlier. In order to evaluate the performance of the adapted LDA, three reference data sets were processed and the results compared with other class-modelling techniques such as SIMCA and UNEQ.