An evaluation of fake fingerprint databases utilizing SVM classification

Abstract Automated fingerprint identification systems are under serious threats from fake fingerprints such as fingerprint films. Fake fingerprint detection is a novel hot topic in fingerprint recognition, and select works covering the detection algorithm are presented. However, the verification of their algorithms is normally based on non-standard fake fingerprint data. This situation degrades the confidence on those detection algorithms, and no fair comparison for their performances is available. In this paper, three public fake fingerprint databases are evaluated by observing the classification accuracies of SVM classifiers with different samples and feature vectors. Through the classifications based on three types of commonly-used fingerprint features (spatial features, detailed ridge features, and Fourier spectrum features), the experimental results show that the fake fingerprints fabricated by latex or body doubles are the most difficult to discriminate, and CASIA and LivDet 13 are the databases used as the standard in the performance verification of fake fingerprint detection algorithms. Additionally, several strong features available for different material and databases are noted.