Analysis of error correcting codes for the secure storage of biometric templates

Error Correcting Codes (ECCs) are widely used today in combination with cryptographic hash functions in the design of biometric systems with secure template storage. The selected ECC strongly influences the performance of the system in terms of False Rejection Rate (FRR) and False Acceptance Rate (FAR) and, therefore, the choice of the code must be done very carefully. This paper presents an analysis of the ECC properties which affect the security and recognition performance of biometric systems, taking as starting point the genuine and impostor normalized Hamming Distance (HD) distributions resulting from intra- and inter-class comparisons. To illustrate the ECC properties relevant for this discussion, the performance of a secure iris verification system is studied. The ECC codes selected for testing are Low-Density Parity Check (LDPC) and Reed Solomon (RS) codes, since these are two of the most commonly used ECCs for secure biometric template storage systems and they have very different properties, which allows testing under a wide range of conditions.