Orthogonal Maximum Margin Projection for Face Recognition

Dimensionality reduction techniques that can introduce low-dimensional feature representation with enhanced discriminatory power are of paramount importance in face recognition. In this paper, a novel subspace learning algorithm called orthogonal maximum margin projection(OMMP) is proposed. The OMMP algorithm is based on the maximum margin projection (MMP), which aims at discovering both geometrical and discriminant structures of the face manifold. First, OMMP considers both the local manifold structure and class label information by using the within-class and between-class graphs, as well as characterizing the separability of different classes with the margin criterion, then OMMP orthogonalizes the basis vectors of the face subspace. Experimental results on three databases show the effectiveness of the proposed OMMP algorithm.