Presentation Attack Detection through Fingerprint Biometric using Leveraged Hard Voting Method

Fingerprint recognition systems have significantly improved in recent years, but they are still vulnerable to spoofing attacks. Traditional fingerprint authentication methods, often based on single-feature and static approaches, may not perform consistently across different sensors or materials used for spoofing. This emphasizes the critical need for an effective fingerprint presentation attack detection (PAD) system to guarantee secure and reliable functionality in practical applications. The risk of fingerprint spoofing underscores the importance of devising reliable techniques to differentiate between genuine and fake fingerprints. Solutions to these vulnerabilities can be implemented using either hardware or software approaches. Hardware-based methods involve gathering additional data about the finger through specialized sensors, such as those measuring spectral absorption, skin temperature, or pulse oximetry. By contrast, software-based methods rely solely on analyzing the digital image captured by the fingerprint sensor, without requiring additional hardware. In this study, we introduce a feature-level approach based on software utilizing a hard voting ensemble that combines random forest (RF), XGBoost, and CatBoost algorithms. Our method outperforms existing techniques in terms of performance. Furthermore, we explore the use of neural networks and discriminant analysis for iris recognition in images. Our approach achieves higher recognition rates and lower computational complexity compared with many machine learning (ML) algorithms like RF, XGBoost and CatBoost. Experimental evaluations using datasets like Clarkson LivDet2015, Digital Persona, CrossMatch, and HighScan yield recognition accuracies of 81.08%, 84.16%, and 92.72%, respectively.