Enhanced ICA Based Face Recognition Using Histogram Equalization and Mirror Image Superposition

Face Recognition (FR) systems accuracy often degrade due to different affecting factors such as varying lighting conditions, expression and pose. The proposed method enhances the accuracy of ICA based FR system presented by Bartlett [8] using novel classification scheme and pre-processing techniques. In this paper, the two architectures are combined using a novel classification scheme that uses the Mode instead of minimum distance as a criterion to recognize faces. In addition, it applies different pre-processing techniques such as Mirror Image Superposition (MIS), Histogram Equalization (HE) and Gaussian Filtering (GF) to overcome pose, expression and lighting variations. MIS is used to neutralize expression and pose variance, HE is used to enhance contrast and Gaussian Filtering (GF) helps in removing noise and thus ensures feature vectors are robust to illumination variations. Experimental results conducted on Yale & LFW database show an increase in recognition accuracy by around 14% for the proposed approach over that of original ICA method by Bartlett N.