A Hybrid Template Protection Approach using Secure Sketch and ANN for Strong Biometric Key Generation with Revocability Guarantee

Nowadays, biometric recognition has been widely applied in various aspects of security applications because of its safety and convenience. However, unlike passwords or tokens, biometric features are naturally noisy and cannot be revoked once they are compromised. Overcoming these two weaknesses is an essential and principal demand. With a hybrid approach, we propose a scheme that combines the Artificial Neural Network (ANN) and the Secure Sketch concept to generate strong keys from a biometric trait while guaranteeing revocability, template protection and noisy tolerance properties. The ANN with high noisy tolerance capacity enhances the recognition by learning the distinct features of a person, assures the revocable and non-invertible properties for the transformed template. The error correction ability of a Secure Sketch concept's construction significantly reduces the false rejection rate for the enroller. To assess the scheme's security, the average remaining entropy is measured on the generated keys. Empirical experiments with standard datasets demonstrate that our scheme is able to achieve a good trade-off between the security and the recognition performance when being applied with the face biometrics.