Towards adept hand-crafted features for ocular biometrics

Abstract

This article presents a hand-crafted feature descriptor for ocular recognition, which as opposed to the deep-learning based approaches, is free from any kind of learning. The proposed approach is able to mitigate the limitations of iris recognition, such as poor iris segmentation, partial or covered iris. The proposed approach leverages by the unique texture present in the periocular region, which can provide complementary details along with the iris modality, or can act as a potential stand alone trait. The proposed descriptor is evaluated on three benchmark databases, namely VISOB, CrossEyed and MICHE. Two of these databases (VISOB and MICHE) provide eye images captured through the smartphones, whereas the third database provides standard eye images registered in visible as well as near-infrared wavelengths. Hence, the evaluation reported in this article becomes a comprehensive one. The experimental results exhibit that the proposed approach proves to be suitable in challenging evaluation frameworks.