Robust retina-based person authentication using the sparse classifier

We address the problem of person authentication, including verification and identification, using the vascular network of the retina. We propose a novel feature extraction process that includes the segmentation of feature points related to anatomical characteristics of the retinal vessel-network, the description of these points with the help of the scale-invariant feature transform (SIFT) and the computation of a final feature vector related to the statistical characteristics of the SIFT-based description. After feature extraction, authentication is conducted with the help of the sparse classifier. We successfully test our methods on two databases, one publicly available and the other one (that we now make available as well) specially generated for this purpose. The results show that apart from high accuracy, the proposed algorithm enjoys a set of invariance properties that make it robust to a set of issues afflicting retina-based person authentication systems, while being fast enough to allow practical deployment.