Abstract

In the past two decads, iris spoofing detection has occupied an ample space in the literature of iris biometics. The textured lens may be used to spoof the Iris Recognition (IR) system by exploiting its external texture. Besides, the soft lens may cause an upsurge in the false rejection rate as it blurs the iris texture. Therefore, it is foremost to identify contact lens in human eyes before accessing an IR system. This paper proposes a novel fusion-based approach to discriminate live iris from contact lens images that combines handcrafted and data-driven features. It also demonstrates a Densely-connected Contact-lens Classification Network (DCCNet) as a data-driven model that is basically a customized Densenet121 framework. The DCCNet features are -pooled with handcrafted counterparts to create a combined feature set. However, the optimal features are identified by top-k feature selection using the Friedman test and are fused through score-level fusion. The assessment of the proposed approach includes several experiments simulated on three iris databases, i.e. Notre Dame (ND) Contact Lens 2013, IIIT-Delhi Contact Lens (IIITD), and Clarkson Databases. The equal error rate (EER) and the detection error tradeoff (DET) curve are used as performance metrics. Further, the statistical analysis is performed using Nemenyi and Bonferroni-Dunn tests, where the proposed approach significantly improves the state of the arts.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call