Face verification using sparse representations

We propose a face verification framework using sparse representations that integrates two ways of employing sparsity. Given an image pair (A, B) and a dictionary D, for image A(B), we generate two sparse codes, one by using the original dictionary and the other by adding B(A) into D as an augmented dictionary. Then the correlation of the sparse codes of A and B, both under the original dictionary D, measuring how similar the pair is, is referred to as the similarity score. The dissimilarity of the sparse codes of A(B), respectively under D and D+B(A), is referred to as the dissimilarity score. We exploit multiple feature transforms to obtain several scores using these two measures and fuse them by simple averaging for the situation where no training set is available or by an SVM when a training set is given. We evaluate our algorithm on the LFW dataset, where it is shown to outperform state-of-the-art methods in the unsupervised setting by a large margin and delivers very comparable performance to methods in the image restricted setting despite its simplicity.