Face representation based on the multiple-class maximum scatter difference

In this paper we extend the maximum scatter difference discriminant criterion which is proposed for binary classification to the multiple-class maximum scatter difference discriminant criterion. Based on this new criterion we establish a novel face representation method. The facial feature extraction method based on the multiple-class maximum scatter difference discriminant criterion effectively avoids the small sample size problem which always brings troubles to conventional discriminant analysis methods when they are applied to face recognition tasks. Experimental results conducted on international benchmark datasets such as ORL, Yale, and FERET face image databases demonstrate that the novel face representation method is promising in comparison with Fisherfaces, eigenfaces, orthogonal complimentary space method, and null space method.