Biometric Detection with Enhanced Security using Federated Leaming

Biometrics refers to the identification and verification of individuals based on unique biological characteristics that are difficult to replicate or forge. With rapid technological advancements, biometric recognition systems have become widely adopted in real-world applications such as fingerprint-based smartphone unlocking, facial recognition for secure building access, and voice authentication for online services. Despite these advantages, traditional biometric systems rely on centralized data storage, which poses serious security risks by creating single points of failure and increasing vulnerability to malicious attacks. In addition to security concerns, training biometric models on large-scale datasets is computationally expensive and complex. Federated learning addresses these challenges by decentralizing the learning process, allowing models to be trained across multiple devices or organizations without transferring raw data to a central server. This approach significantly enhances data privacy and security while reducing computational overhead. By enabling organizations to retain full ownership and control of sensitive biometric data, federated learning makes it possible to leverage deep learning models without compromising confidentiality. This project presents a federated learning-based framework for biometric fingerprint detection using a ResNet deep learning model. ResNet's capability to efficiently train deep architectures makes it highly suitable for extracting hierarchical and complex fingerprint features. The model is trained on the SOCOFing dataset, which contains 6,000 fingerprint images from 600 African subjects, with ten fingerprints per individual, all aged 18 years or above. The dataset includes detailed labels such as gender, hand, and finger type, along with synthetically altered fingerprint images featuring varying levels of obliteration, central rotation, and z-cut alterations, enabling robust evaluation of the proposed system.