Feature Transformation of Biometric Templates for Secure Biometric Systems Based on Error Correcting Codes

Abstract

Secure storage of biometric templates is extremely important because a compromised biometric cannot be revoked and replaced an unlimited number of times. In many approaches proposed for secure biometric storage, an error correcting code (ECC) is applied to the enrollment biometric and the resulting parity or syndrome symbols are stored on the access control device, instead of the original biometric. The principal challenge here is that most standard ECCs are designed for memoryless channel statistics, whereas the variations between enrollment and probe biometrics have significant spatial correlation. To address this challenge, we propose to transform the original biometric into a feature vector that is explicitly matched to standard ECCs, thereby improving the security-robustness tradeoff of the overall biometric system. As a concrete example, we transform fingerprint minutiae maps into feature vectors compatible with ECCs designed for a binary symmetric channel. We conduct a statistical analysis of these feature vectors and show how our feature transformation algorithm may be combined with Low-Density Parity Check (LDPC) codes to obtain a secure fingerprint biometric system.