Face recognition using wavelet transforms-based feature extraction and spatial differentiation-based pre-processing

Face Recognition (FR) across pose is a problem of fundamental importance in computer vision. We propose to address this problem using three novel techniques, viz., Spatial Differentiation (SD), Wavelet Transform based Feature Extraction (WTFE), and Twin Pose Testing Scheme (TPTS), to improve the performance of a FR system. SD is used to enhance the facial features. WTFE uses the shift invariance property of SWT which, along with TPTS, neutralizes pose variations. A Binary Particle Swarm Optimization (BPSO) based feature selection algorithm is used to search the feature space for the optimal feature subset. Individual stages of the FR system are examined and an attempt is made to improve each stage. Experimental results, obtained by applying the proposed algorithm on four benchmark face databases, namely, Color FERET, FEI, LFW and IFD, show that the proposed system outperforms other FR systems. A significant increase in the recognition rate and a substantial reduction in the number of features selected are observed.