Random Undersampling and Local-Global Matching Mechanism for Cancellable Biometrics Against Authentication Attack

Ying Zhou, Ming Jie Lee, Hui Zhang, Xingbo Dong, Zhe Jin

Abstract: The trade-off between security and verification performance is inevitable towards biometric template protection. The system developer has to sacrifice some genuine acceptance rate and tune the matching threshold to tolerating more false acceptance. To alleviate this problem, we introduce a new method of feature transformation and matching, which consists of a random undersampling and local-global matching mechanism for the hashing-based cancellable biometrics. This method manages to enlarge the gap between the mean of genuine/ impostor score distributions. As such, the decision environment is improved and the biometric system could provide more resistance to authentication attacks. Comprehensive experiments are conducted on the fingerprint FVC2002 and FVC2004 datasets, and the results demonstrate that the proposed method improves the decision environment in terms of decidability and verification performance.