Dynamic Fingerprint Statistics: Application in Presentation Attack Detection

Abstract

Fingerprint recognition systems have proven significant performance in many services such as forensics, border control, and mobile applications. Even though fingerprint systems have shown high accuracy and user acceptance, concerns have raised questions about the possibility of having our fingerprint pattern stolen and presented to the system by an imposter. In this paper, we propose a dynamic presentation attack detection mechanism that seeks to mitigate presentation attacks. The adopted mechanism extracts the variation of global fingerprint features in video acquisition scenario and uses it to distinguish bona fide from attack presentations. For that purpose, a dynamic dataset has been collected from 11 independent subjects, 6 fingerprints per user, using thermal and optical sensors. A total of 792 bona fide presentations and 2772 attack presentations are collected. The final PAD subsystem is evaluated based on the standard ISO/. Considering SVM classification and 3 folds cross validation, the obtained error rates at 5% APCER are 18.1% BPCER for the thermal subset and 19.5% BPCER for the optical subset.