Privacy Preserving Fingerprint Classification Using Federated Learning

Fingerprint recognition is a widely used biometric authentication method that faces challenges due to the increasing sophistication of fake fingerprint creation techniques. This work focuses on the task of real and fake fingerprint classification using federated learning, leveraging a custom dataset derived from the CASIA fingerprint dataset. The goal is to develop a model capable of accurately distinguishing between real and fake fingerprints for different clients. The methodology involves training individual client models on their respective data using Convolutional neural network (CNN) architectures. A global model is updated using a federated learning technique that ensures collaborative model training while protecting data privacy. Using a separate test dataset, the performance of the trained global model is assessed, and several performance metrics, including accuracy, precision, recall, and F1-score, are produced. The results demonstrate the effectiveness of federated learning in addressing the real and fake fingerprint classification task. The custom dataset derived from the CASIA fingerprint dataset provides a focused analysis of the binary classification problem, enabling insights into the model’s ability to accurately classify real fingerprints while minimizing false positives and false negatives. The findings contribute to the development of robust fingerprint recognition systems and highlight the potential of federated learning in addressing biometric authentication challenges.