Fingerprint Multiple-Class Classifier: Performance Evaluation on Known and Unknown Fingerprint Spoofing Materials

Fingerprint recognition is a popular and reliable biometric technology used in many security-sensitive applications. However, the use of fake fingerprints made from ubiquitous spoofing materials poses a significant threat to security systems. While several studies have proposed binary classifiers to detect fingerprint presentation attacks, relatively few have explored the effectiveness of multiple-class classifiers in detecting known and unknown spoofs. In this study, we evaluated the efficacy of multiple-class classifiers using deep transfer learning to detect presentation attacks made with different spoofing materials. Our experiments on the LivDet 2009-2015 datasets showed that while a classifier model developed without data augmentation performed better on known spoofs, it showed poor performance on cross-material detection of all seven fingerprint spoofing materials. These results suggest that modelling a multiple-class classifier is not an efficient approach for detecting cross-material presentation attacks in fingerprint recognition systems.