A New Face Recognition Framework: Symmetrical Bilateral 2DPLS plus LDA

Abstract—A novel face recognition framework is proposed in this paper to alleviate "Small Sample Size " (SSS) problem of the conventional Linear Discriminant Analysis (LDA). This method is based on the feature extraction of global odd and even face image representation, and a dimension reduction process via Symmetrical Bilateral 2D Partial Least Square Analysis(2DPLS). The low-dimensional features are then used to train a LDA classifier which uses Frobenius-norm classification measure, and uses pseudo inverse to make sure between-class matrix Sw be full rank. Experimental results on Yale Face Database B, ORL, and FERET Face Database demonstrate that our framework is highly efficient and gives the state-of-the-art recognition rate. Index Terms—face recognition, pseudo inverse, Frobenius-