Gabor Feature-Based Face Recognition Using Riemannian Manifold Learning

This paper introduces a novel Gabor-based Riemannian manifold learning (GRML) method for face recognition. Riemannian manifold learning (RML) is a recently proposed framework which formulates the dimensionality reduction of a set of unorganized data points as constructing normal coordinate charts for an underlying Riemannian Manifold. In this paper, we investigate its practical version for face recognition,which is characterized both by the selection of the coordinate chart and by the out-of-sample testing. The GRML method applies the RML to an augmented Gabor feature vector derived from the Gabor wavelet representation of face images. Experiments show that compared with Gabor-based PCA (GPCA), our GRML achieves better recognition performance.