Abstract:

It is well-known that a biometric fuzzy vault can be constructed by applying an error correcting code (ECC) to a biometric signal. This is attractive because authentication only requires the check bits of the ECC to be stored on the access control device, whereas the personal biometric traits need not be stored. For a given coding rate, the ECC attempts to correct the errors between an enrollment biometric and the provided probe, and authenticates if it is successful in doing so. Unfortunately, most implementations of biometric fuzzy vaults have very poor robustness to the inherent noisiness of biometric measurements. In this paper, we provide ECC design considerations for secure biometric systems, which provide both better robustness and greater security. In particular, for any feature extraction algorithm, we propose to reorder the feature bits according to their reliability, and associate the reliable bits with high-degree variable nodes in the graph of the ECC. Further, the reliability of a bit is measured at enrollment and used to initialize the ECC decoding. Experiments on an extensive database show considerable reduction in the false reject rate, while restricting the successful attack rate to a very low value.