Robust face recognition under lighting variations

In this paper, we propose a new face recognition algorithm based on the matching of relative image gradient magnitudes between images. The recognition algorithm first uses a face localization procedure to provide rough face regions under different lighting conditions, followed by an iterative optimization procedure for precise face matching. Both our face localization and matching procedures are based on matching relative image gradient to be robust against lighting variations. Then a robust face similarity measure based on comparison of relative image gradients is used to determine the face recognition results. The face localization step finds some candidate poses of the face in the image through a fast k-NN search of the best match of the relative gradient features from the database of training feature vectors, which are obtained through image synthesis. After the face images are aligned, the face similarity measure is computed from the normalized correlation between the relative gradients. Experimental results are shown to demonstrate its robust recognition performance under different lighting conditions.