Evaluation of candidate fingerprint features for employment within template-free biometric cryptosystems

With the increasing employment of biometric authentication worldwide, there have been significant concerns over the security and privacy of the biometric templates, which are stored and used for matching. This thesis investigates a relatively new area of biometric authentication, known as biometric encryption. Biometric encryption is an alternative form of authentication where an encryption key is either generated, or retrieved, instead of a binary result indicating verification. The key can subsequently be used to encrypt and decrypt data using existing encryption algorithms, such as triple-DES, AES and RSA. The purpose of this thesis is to investigate the feasibility of generating stable and strong encryption keys directly from biometric features, without the use of templates. To accomplish this, a proof-of-concept was created using the fingerprint biometric modality. Fingerprints currently possess a limited range of features, although they are highly variable and result in many user distributions overlapping. Therefore, if consistent, highly entropic keys can be generated using fingerprint features, other modalities, whose feature distributions are better behaved, should improve the performance of the generated key. By performing statistical tests, on the countless variants of existing and novel feature vectors, the best performing feature vectors were then selected for deployment in a template-free biometric encryption system. Although the results were not exceptional, an encryption key, with 21 effective bits, was reproduced correctly -80% of the time with 55% unique keys in the resultant key-space. These results are encouraging, although significant improvements would be necessary in order to obtain the performance required of a practical system.