Conventional Neural Network-Based Detection of Fingerprint Spoofing Attacks

Abstract

Designing a secure, autonomous, and adaptive open-set framework is essential for reliably determining whether a fingerprint input is from a live source. One of the key challenges in high security systems is the misclassification of spoof fingerprints as genuine, known as a Type-I error. This issue becomes more pronounced due to the limited availability of spoof samples. To address this limitation, we propose a novel fingerprint presentation attack detection (FPAD) method that exclusively utilizes genuine (live) fingerprint samples. Our technique begins with acquiring a high-quality fingerprint sample from the authorized individual. From each live sample, six image quality metrics are derived, which together constitute what this set of six quality-based metrics. These are collectively recognized as the transient liveness factor (TLF). These metrics serve as adaptive indicators of authenticity. To ensure robust validation, the method employs a fusion approach that incorporates three anomaly detection techniques: one-class CNN, isolation forest, and local outlier factor. The system demonstrates exceptional spoof detection performance, achieving 100% accuracy under open-set conditions. Furthermore, we explore the integration of this approach into cloud-based environments and discuss the ongoing challenges in implementing user-specific spoof detection systems.