Two-dimensional direct and weighted linear discriminant analysis for face recognition

In this paper, a novel algorithm for feature extraction—two-dimensional direct and weighted linear discriminant analysis (2D-DWLDA)—is proposed. The improvement of 2D-DWLDA algorithm over traditional linear discriminant analysis (LDA) and 2D-LDA methods benefits mostly from three aspects: (1) 2D-DWLDA is based on 2D image matrices rather than 1D vectors, so the scatter matrices can be constructed directly using the image matrices, and calculated accurately; (2) by introducing weighting function, the overlap of the neighboring classes is weaken; (3) direct LDA method is utilized so that the extracted features have more discriminant power. Finally, we performed a series of experiments on three face databases: ORL, CAS-PEAL and Yale database, the recognition accuracies are higher using 2D-DWLDA than 2D-LDA and LDA.