An Effective Orchestration for Fingerprint Presentation Attack Detection

Abstract

Fingerprint presentation attack detection has become significant due to a wide-spread usage of fingerprint authentication systems. Well-replicated fingerprints easily spoof the authentication systems because their captured images do not differ from those of genuine fingerprints in general. While a number of techniques have focused on fingerprint presentation attack detection, they suffer from inaccuracy in determining the liveness of fingerprints and performance degradation on unknown types of fingerprints. To address existing limitations, we present a robust fingerprint presentation attack detection method that orchestrates different types of neural networks by incorporating a triangular normalization method. Our method has been evaluated on a public benchmark comprising 13,000 images with five different fake materials. The evaluation exhibited our method’s higher accuracy in determining the liveness of fingerprints as well as better generalization performance on different types of fingerprints compared to existing techniques.