Fingerprint Reconstruction from Biometric TemplatesMijatović, Aleksandar (Avtor) Peer, Peter (Mentor) Oblak, Tim (Komentor) fingerprint reconstructionminutiae encodingpix2pixModern fingerprint recognition systems rely heavily on compact minutiae-based templates for efficient processing of biometric data from our fingertips. While minutiae templates were originally assumed to be non-reversible, recent studies have demonstrated that realistic fingerprint images can be reconstructed from these sparse representations. Despite significant progress, existing generative approaches often struggle with recreating the complex ridge structure of fingerprints. Reconstructed impressions frequently exhibit artifacts near singular regions, disrupted global ridge flow, and spurious minutiae, particularly in areas lacking sufficient contextual information. In this work, we introduce a context-aware minutiae encoding that provides the generative model with additional structural information during the training process. The proposed approach improves the global ridge continuity, reduces artifacts near singularities, and generates more natural looking ridges. Evaluation was performed on URU and Anguli datasets using biometric matching-based attack scenarios and NFIQ2 quality assessment. Averaged across three thresholds, Type-I attack performance improved by 17% on the URU dataset and 65% on the Anguli dataset, while Type-II attack performance improved by 25% on URU. Furthermore, the average quality of impressions improved by approximately 15--20% on both datasets. These results demonstrate that incorporating richer contextual information significantly enhances reconstruction fidelity, producing higher-quality and more realistic looking fingerprints.20262026-03-25 10:10:03Diplomsko delo/naloga181103VisID: 38154COBISS_ID: 275592195sl