An Approach to Iris Contact Lens Detection Based on Deep Image Representations

摘要: Spoofing detection is a challenging task in biometric systems, when differentiating illegitimate users from genuine ones. Although iris scans are far more inclusive than fingerprints, and also precise for person authentication, recognition systems vulnerable to spoofing via textured cosmetic contact lenses. Iris referred as liveness (binary classification of fake real images). In this work, we focus on three-class problem: images with (colored) lenses, soft no Our approach uses convolutional network build deep image representation an additional fully-connected single layer max regression classification. Experiments conducted comparison state-of-the-art (SOTA) two public databases lens detection: 2013 Notre Dame IIIT-Delhi. can achieve 30% performance gain over SOTA the former database (from 80% 86%) comparable results latter. Since IIIT-Delhi does not provide segmented and, differently SOTA, our segment yet, conclude that these very promising results.