An extreme case of the generalized optimal discriminant transformation and its application to face recognition

A study has been made on an extreme case of generalized optimal set of discriminant vectors. Equivalence between the generalized K-L transformation and the generalized optimal discriminant transformation is proved under the condition that the population scatter matrix of training samples is nonsingular. A new algorithm for determining the generalized optimal set of discriminant vectors is proposed based on the above theory, which is applied to the feature extraction of human face images. The results of experiments conducted on ORL and Yale databases show the effectiveness of the new feature extraction algorithm based on the extreme case of the generalized optimal discriminant transformation.