On applicability of Principal Component Analysis to concept learning from images

Abstract

The paper presents some experiments investigating the applicability of the Principal Component Analysis method for solving several concept learning tasks defined on images of faces. The results have shown that, in most cases, the applied transformation improves the classification accuracy of used concept learning algorithms. In addition the experiments have confirmed a possible relation between the quality of the obtained improvements and the complexity of the concepts to be learnt. This relation has the potential to be an objective measure of “concept complexity”.