Orthogonal Neighborhood Preserving Discriminant Projections for Face Recognition

Feature extraction is an important step for face recognition. The capability of feature extraction directly influences the performance of face recognition. Recently, some manifold learning algorithms have drawn much attention. Among them, neighborhood preserving projections is one of the most promising feature extraction techniques. Though NPP has been applied in many fields, it has limitations to solve recognition task. In this paper, a novel feature extraction method called orthogonal neighborhood preserving discriminant projections (ONPDP) is proposed. The aim of ONPDP is to preserve the within-class neighboring geometric relation by taking into account class label information, while maximizing the between-class distance. Two abilities of manifold learning and classification are combined into the properties of the proposed algorithm. In addition, ONPDP orthogonalizes the set of the projection transformation vectors of the face subspace by Gram-Schmidt orthogonalization and thus the stronger discriminating power can be obtained. Experiment results on FERET and ORL face databases demonstrate the effectiveness of the proposed method.