Privacy-Secure and Decentralized Biometric Authentication Models Using Federated Learning Frameworks

A biometric identifier, in contrast to other identifiers used for authentication, is a quantitative evaluation of an individual's physical attributes that is successfully used to confirm or validate the identification. Because of its remarkable and consistent texture variation, iris recognition is thought to be the most dependable biometric recognition. Concerns about security, authentication, and identification are growing in importance across all domains as technology advances daily. High-security applications use these unique patterns for iris recognition. To increase the security of biometric applications without jeopardizing individual privacy, this research suggests a novel FL technique for privacy-preserving iris detection. Traditional biometric systems rely on centralised data storage and processing, which threatens data security and misuse. Federated Learning: decentralized model training that allows multiple federated clients (like mobile devices or edge servers) to collaboratively train a global model without sharing their raw iris data. We explore how this architecture can be applied to iris detection systems, maintaining accuracy while mitigating privacy risks. Our method ensures that iris features are learned locally on individual devices, and only encrypted model updates are shared with the central server. To strengthen privacy, we incorporate differential privacy and secure aggregation techniques to prevent data leakage during model training. Experimental results demonstrate the effectiveness of the proposed framework in maintaining high detection accuracy, low communication overhead, and robust privacy guarantees.