Fingerprint Restoration and Identification using Pix2Pix cGAN and Triplet Loss

Fingerprint restoration and identification is a critical biometric authentication method with security and digital identity verification applications. Its accuracy relies on image quality and identification algorithms. Real-world fingerprint images often suffer from noise and distortions. This work presents a novel approach using a Pix2Pix Conditional GAN (CGAN) for fingerprint restoration. We evaluated the method using the triplet loss function by measuring Euclidean distances between anchor, positive, and negative embeddings. Rigorous testing across various image qualities enhances network performance. Combining CGAN with the Siamese triplet network, shown in this study, leads to substantial improvements in fingerprint restoration and identification results. We achieved the best accuracy of 86% in fingerprint identification while preserving the quality of the dataset designed for this work. This innovation holds potential for real-world applications, notably in crime scene investigation and forensic sciences, where recovering and enhancing fingerprint data are crucial.