A Multi-Attack Iris Spoof Detection Framework: An Enhanced CNN-Siamese Network
TL;DR
A CNN-based Siamese network for detecting iris spoofing using genuine iris images and simulated attack variations is proposed, supporting the strong discrimination capability between genuine and spoofed image specifically on simulation the frequency based synthetic iris-attack instruments.
Abstract
Iris spoof detection is essential for protecting biometric recognition systems against presentation attacks such as printed images, contact lenses, and digitally generated iris patterns. This work proposes a CNN-based Siamese network for detecting iris spoofing using genuine iris images and simulated attack variations. The model learns discriminative representations from pairs of iris images and determines whether they belong to the same or different classes using pairwise distance. Multiple spoofing scenarios, including print-like appearance, contact-lens effects, and frequency-spectrum manipulation, are incorporated to evaluate the robustness of the approach. The proposed framework reports strong and consistent performance of on IIT Delhi iris dataset, across all accounted metrics; accuracy, precision, recall, and F1-score. Further, it reports an AUC of 1.00 supporting the strong discrimination capability between genuine and spoofed image specifically on simulation the frequency based synthetic iris-attack instruments. The proposed method is also compared with a few baseline existing methods for analyzing its performance and computational complexity under resource-limited enviornment.