Fingerprint Presentation Attack Detector Using Global-Local Model.

Area of Science:

• Biometrics
• Computer Vision
• Security Systems

Background:

• Automated fingerprint recognition systems (AFRSs) are vulnerable to presentation attacks (PAs).
• Existing presentation attack detection (PAD) methods suffer from information loss and poor generalization.
• New PA materials and fingerprint sensors pose ongoing challenges for PAD.

Purpose of the Study:

• To propose a novel global-local model-based PAD method (RTK-PAD).
• To overcome limitations of information loss and poor generalization in current PAD techniques.
• To improve the accuracy and robustness of fingerprint spoof detection.

Main Methods:

• A global module using cut-out for nonlocal features and a global spoofness score.
• A local module employing texture in-painting for fingerprint patch analysis and a local spoofness score.
• A rethinking module connecting global and local modules by localizing discriminative patches.
• Fusion of global and local spoofness scores for final PAD decision.

Main Results:

• The RTK-PAD method achieved an average classification error (ACE) of 2.28% on LivDet 2017.
• It attained a true detection rate (TDR) of 91.19% at a 1.0% false detection rate (FDR).
• RTK-PAD significantly outperformed state-of-the-art methods, showing a ~10% improvement in TDR.

Conclusions:

• The proposed RTK-PAD method effectively addresses limitations in existing PAD techniques.
• The global-local model demonstrates superior performance in detecting presentation attacks.
• RTK-PAD offers a more robust and generalizable solution for fingerprint security.