A realtime fingerprint liveness detection method for fingerprint authentication systems

Abstract

Fingerprint authentication is widely used in various intelligent devices. Fraudulent Attack (FA) using forged fingerprints is one of the most seen security threats in fingerprint authentication. Fingerprint Liveness Detection (FLD), a system to detect if a fingerprint is natural (from a live person present at the point of capture) or fake (from a spoof artifact or lifeless body part), is an essential step before fingerprint authentication. This paper proposes a lightweight and real-time FLD method based on a comprehensive learning system with a Uniform Local Binary Pattern (ULBP) for Fingerprint Authentication Systems (FAS). Our approach contains three steps: First, perform the Region of Interest (ROI) extracts for fingerprint samples to eliminate data noise. Second, to construct distinguishable texture features by ULBP descriptors as the Broad Learning System (BLS) input. ULBP reduces the variety of binary patterns of fingerprint features without losing any critical information. Third, the extracted features are fed into the BLS for subsequent training. The BLS is a flat network that transfers and places the original input as a mapped feature in feature nodes, generalizing the structure in augmentation nodes. The results from the experiment with the five public fingerprint datasets (LivDet 2011, 2013, 2015, 2017, and 2019) show that, compared to other advanced models, our method has faster speed, smaller size, and higher performance. Our method can achieve excellent FLD results in non-high-performance equipment, which is critical for its application in mobile devices.