Towards a biometric random number generator – a general approach for true random extraction from biometric samples

Abstract

Biometric systems are per definition used to identify individuals or verify an identity claim - one difficulty of getting reliable decisions is the inherent noise that makes it difficult to extract stable features from biometric data. This paper describes how biometric samples can be used to generate strong random numbers which form the basis of many security protocols. Independent from the biometric modality, the only requirement of the proposed solution are feature vectors of fixed length and structure. Each element of such a feature vector is analyzed for its reliability - only unreliable positions, that cannot be reproduced coherently from one source, are extracted as bits to form the final random bit sequences. Optionally a strong hash-based random extraction can be used. The practicability is shown testing vascular patterns against the NIST-recommended test suite for random number generators.