IVIDNet: Intelligent iris vitality detection via weighted prediction score level fusion

Abstract

Iris recognition is one of the most widely used human authentication mechanism that has gained huge popularity due to its security and efficiency. However, recently attackers have breached the security of these systems via synthetically generated fake iris traits using a variety of artefacts such as paper printouts, video attacks, and cosmetic lenses. To countermeasure these attacks, an anti-deception sub-system is instilled between sensor and feature extractor modules of the system. These sub-systems intelligently measure the vitality of the presented iris traits and are popularly known as iris vitality detection (IVID) or iris anti-spoofing techniques. Recently, the deep learning-based models are widely deployed for iris anti-spoofing approaches, but these methods result in additional training overhead as well as require larger training dataset. In this paper, we expound a novel iris vitality detection network (IVIDNet) that employs the robust features of two pre-trained deep convolutional neural networks (DCNN). The extracted features are then integrated via a new weighted score-level fusion technique. To demonstrates the efficacy of the proposed technique, experimental analysis is carried out on benchmark iris liveness detection Notre Dame 2017 and Notre Dame cosmetic lenses dataset (NDCLD) 2015. The proposed IVIDNet exhibits excellent performance in the known attack scenario with an average classification error rate (ACER) of 0.60%. Moreover, IVIDNet shows superior performance in the unseen attacks scenario (i.e. cross-database) with an ACER of 9.72%. In overall, the proposed model outperforms other similar iris anti-spoofing approaches and efficient in terms of computational overhead. © 2023, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.