Combination of DFT as Global Face Descriptor and LBP/LDiP/LDNP as Local Face Descriptor for Face Recognition

Abstract

This paper describes the combination of DFT as a global face descriptor and LBP/LDiP/LDNP as a local face descriptor that results in a final feature vector. Each of these face descriptors does not need a complex learner to classify a novel face pattern when operates separately. However, it will not work when they combine together. The main contributions of our work are in determining the final feature vector that discriminatively represents a face image and the optimal classifier (SVM) that efficiently and accurately classify a novel feature pattern. We conduct simulations on ORL face database by varying the number of face images in training and testing sets on two well-known global face descriptors (PCA and LDA), three local face descriptors (LBP, LDiP, and LDNP), and also the combination of DFT and LBP/LDiP/LDNP. Simulation results show that, the more the number of face images in the training phase, the better the recognition rate of the combination face descriptors rather than either each global or local face descriptor.