State-Of-The-Art Machine Learning and Deep Learning Techniques in Iris Recognition: A Review

Iris recognition has advanced remarkably from traditional handcrafted techniques to powerful deep learning-driven solutions. This review synthesizes findings from over 30 high-impact studies to examine the evolution of iris recognition methodologies, with a focus on machine learning (ML), deep learning (DL), and hybrid frameworks. Innovations in convolutional neural networks, transformer models, generative adversarial networks, and transfer learning have enabled high accuracy (≥99%) in real-time, cross-spectral, and post-mortem scenarios. Key advancements include YOLOv4-tiny–Efficient Net pairings for edge deployment, VGG–ResNet ensembles for spoof detection, and saliency-guided training for explainability. Foundation models like DinoV2 further enhance cross-domain generalization. Alongside technical improvements, the field has emphasized privacy through federated learning and cancelable biometrics. The review also highlights diverse applications—from PAD and liveness detection to forensic analysis—underscoring a shift toward robust, interpretable, and secure biometric systems. Overall, deep learning has redefined the iris recognition landscape, offering scalable solutions adaptable to modern security and identification needs.