Semi-supervised Discriminative Orthogonal Locality Preserving Projections for Face Recognition

Locality Preserving Projections (LPP) has been a popular method for feature extraction techniques. However, when applied to classification problems in a supervised setting, LPP only focuses on the intraclass geometrical information while ignores the interaction of samples from different classes. To enhance the performance of LPP in classification, a new algorithm termed as Semi-supervised Discriminative Orthogonal locality preserving projections (SDOLPP) is proposed in this paper. SDOLPP takes into account the labeled and unlabeled samples, changes the objective function, and then orthogonalizes the basis vectors of the face subspace. The proposed method was compared with LPP, DOLPP and SLPP on the AR and YaleB face databases. Experimental results verify the performance of the proposed approach.