Deep Learning for Biometric Data Augmentation: An Attention-Based GAN Approach for Fingerprints

Abstract

The advancement of biometric security systems has placed significant emphasis on the reliability, scalability, and privacy of fingerprint recognition technologies. However, the limited availability of diverse and high-quality fingerprint datasets continues to be a major bottleneck in training robust machine learning models. This research proposes a deep learning-based framework for synthetic fingerprint data augmentation using an attention-guided Generative Adversarial Network (GAN). The proposed model is capable of generating realistic and varied fingerprint images by learning intrinsic fingerprint patterns through adversarial training while integrating attention mechanisms for enhanced structural detail retention. By incorporating self-attention modules in both generator and discriminator networks, the model learns contextual dependencies and structural coherence across fingerprint ridges and minutiae. Experiments are conducted on publicly available datasets such as FVC2004 and PolyU HRF, demonstrating that the synthetic fingerprints not only maintain high visual fidelity but also significantly improve the performance of downstream fingerprint classification and matching tasks. This work contributes to addressing data scarcity challenges and opens new avenues for privacy-preserving biometric model development.