Work remains on fingerprint biometric presentation attack detection despite promising stats
Work on detecting presentation attacks against fingerprint biometric systems is continuing to advance, but even measuring the success of current methods is not entirely consistent across the field, according to presenters at a two-day online event organized by the European Association for Biometrics (EAB) and led by EAB’s Dinusha Frings and Marta Gomez-Barrero of Hochschule Ansbach.
Fingerprint presentation attack detection (PAD) techniques based on deep learning algorithms and other models, as well as additional sensors such as for pulse or illumination were discussed. Real incidents of presentation attacks were reviewed, and the development of PAD evaluation methods and standards were considered.
The event kicked off with a primer on the subject by Gomez-Barrero, and a tutorial highlighting new concepts and metrics in the field of fingerprint biometric presentation attack detection.
Some of the models and methods presented can be used for PAD with different biometric modalities with little adaptation, and models within a single modality are often found to be significantly less effective for some attack types, such as transparent fingerprint overlays, than for others, like fake fingers.
Although some fingerprint PAD mechanisms have shown accuracy above 99 percent, a set of problems remain to be solved. Generalizing for unknown attack types and instruments and testing in realistic scenarios with highly skilled attackers continue to pose challenges.
Presentations on the first day focused on issues related to PAD on mobile device fingerprint scanners, the suitability of finger impedance to presentation attack detection, and on the second day the segmentation of fingerprint and sweat ducts in volumetric data and the practical implications of the ISO/IEC 30107 standard. Stephanie Schuckers of Clarkson University gave a presentation on FIDO certification and PAD, and Gomez-Barrero moderated a panel discussion with representatives from governments, academia, Secunet and Jenetric on the state of the art.
Impedance, like several approaches to fingerprint PAD, shows promise, but requires further refinement to advance the security of fingerprint biometrics against presentation attacks.
The final panel discussed how closely academic and industry work on PAD reflects the real threat posed by expert attackers, and whether multi-modal biometric systems can be a substitute for PAD technology.
The EAB will hold an event on fingerprint image quality assessment in collaboration with NIST and eu-LISA in June.