Automated face recognition using adaptive subspace method

Automated face recognition is reemerging as an active research area because of its various commercial and law enforcement applications. In this paper, we propose a novel approach called the adaptive subspace method motivated by the traditional eigenfaces approach. Our scheme begins with the standardization of face images in order to achieve some invariance of face representation under different image acquisition conditions. Then we combine the K-L expansion technique with genetic algorithms to construct an optimal feature subspace for identification. Finally, any input face image can be projected into this adaptive subspace to be identified using a minimum distance classifier. Experimental results are also given in detail and show our approach offers superior performance.