Palm vein template protection scheme for resisting similarity attack

Abstract

Palm vein recognition technology has rapidly developed due to its high confidentiality and the advantages of liveness detection. Various palm vein template protection methods have emerged to safeguard palm vein data from theft and attack. To fulfill the performance loss requirement of the ideal biometric template protection scheme, these methods make the distribution of the original palm vein data and the palm vein protection templates have a strong distance preserving property (similarity preserving), making it difficult to defend against similarity attack (SA). To address these risks and prevent palm vein data leakage, we propose a Nonlinear Spectral Hashing (NSH) method for palm vein template protection. To obtain palm vein templates with both performance and security, the method first performs random projection on palm vein data to obtain revocable and unlinkable palm vein features. Subsequently, through spectral graph partitioning, it achieves mapping with a preserved similarity structure for palm veins, avoiding excessive performance loss. The method then employs a nonlinear activation function to alter the distribution of palm vein templates with large post-mapping differences, resulting in a uniform distribution of inter-class distances for palm vein data. By reducing the distance-preserving properties, the method enhances protection against similarity attacks. Finally, a sign function is applied to obtain non-invertible binary palm vein templates. Experimental evaluations on the public Tongji University palm vein database assess the method's non-invertibility, revocability, unlinkability, resistance to similarity attack, and recognition performance. The results indicate an Equal Error Rate (EER) of 0.50 %, demonstrating the method's ability to maintain good recognition performance while ensuring high security.