Biometric Key Generation with a Parametric Linear Classifier

Biometrics play an important role in modern access control systems to overcome the problems of forgotten, stolen or easily guessed passwords. With the recent developments in cryptography, hashing and random number generation, biometrics and cryptography are combined in new generation crptosystems. In this paper, it is aimed to produce a unique binary biometric identity code (bit string) by using the individual's fingerprint features and this bit string is then hardened by an error correction scheme. In the proposed scheme sectoral spectral features are extracted from a data set of 702 fingerprints, and then these samples are classified with a parametric linear classifier Each class in this space is represented with a Gauss distribution whose parameters are then converted to a bit string after a fuzzy mapping scheme is applied. The resulting biometric identity code is then strengthened with error correction. If the proposed method is improved the obtained final bit string might be used directly either for authentication or as a unique seed for biometric key generation. The major contribution of this study is to propose a bit generation scheme not requiring a synthetic information such as an output of a pseudorandom or truly random number generator