Touchless Fingerprint Recognition in the Post-COVID Era: A Machine-Learning-Centric Review of Features, Matching, Classification and Security

Abstract

Touchless fingerprint recognition removes physical contact from fingerprint capture through cameras or other optical sensors while computationally recovering ridge information. COVID-19 did not create the technology, but it made the hygiene and shared-surface limitations of contact-based biometrics more visible. This review emphasizes acquisition, segmentation, enhancement, pose normalization, features, matching, classification and presentation-attack detection (PAD). Reported work shows a wide performance range: HOG + SVM reached 100% in one touchless classification experiment, while HOG + decision tree reached 67%; transfer learning, Siamese learning and hybrid generative approaches have reported results above 93% on their respective datasets. These values are not a common benchmark. Compared with face, iris and voice, touchless fingerprint offers a useful balance of permanence, distinctiveness, mature infrastructure and hygienic capture, while challenges remain in pose, illumination, cross-sensor interoperability, spoof resistance, privacy, dataset diversity and edge efficiency.