A Robust CNN-Siamese Framework for Iris Deepfake Spoof Detection with Superior Accuracy and AUC

Abstract

The increasing sophistication of spoofing attacks poses a significant threat to the reliability of iris recognition systems, especially with the rise of deepfake-generated synthetic irises. This study proposes A Robust CNN-Siamese Framework for Iris Deepfake Spoof Detection with Superior Accuracy and AUC, aimed at effectively distinguishing between real and fake iris images. The framework combines a Convolutional Neural Network (CNN) for deep feature extraction and a Siamese Network to measure similarity between input and reference samples. By learning discriminative patterns and applying a threshold-based classification strategy, the system excels at identifying spoofing attempts involving printed images, textured contact lenses, and synthetic irises. Extensive experimentation demonstrates the model’s superiority over traditional classifiers, achieving 97.2% accuracy, 96.8% precision, 97.5% recall, and an F1-score of 97.1%. The system also achieves an AUC of 0.99, highlighting its excellent class separation capability. Additional evaluations, including PCA visualization and FAR/FRR analysis, confirm its robustness and generalizability. This work contributes a scalable, efficient, and secure approach for strengthening iris-based biometric authentication systems against evolving deepfake threats.