Towards Robust Combination of Neural Networks for Fingerprint Presentation Attack Detection

Seong-Hee Park, Min Young Lim, Dongwoo Kang, Youn Kyu Lee

Abstract: To resolve security threats on fingerprint authentication systems, a number of fingerprint Presentation Attack Detection (fingerprint PAD) methods have been proposed. However, existing methods still provide limited performance in terms of detection accuracy and generalization performance. In this paper, we propose a new fingerprint PAD method that ensembles Feature-based Neural Network (FNN) and Convolutional Neural Network (CNN) models based on a weighted voting mechanism. We also designed a new FNN architecture and a new CNN architecture, respectively, which provide improved performance for PAD. To verify the effectiveness of our method, we performed experimental evaluations using real-world datasets, LivDet 2015. The results showed that our method provided improved finger-print PAD accuracy compared to existing methods.