Classification Based Revocable Biometric Identity Code Generation

Abstract

The recent biometric template protection methods often propose salting or one-way transformation functions and biometric cryptosystems which are capable of key binding or key generation to provide the revocability of the templates. Moreover, the use of multiple instances of a biometric trait proposes more robust features which are then combined with the well-known template protection methods. In this study, a normal densities based linear classifier is proposed to distinguish the features associated with each user and cluster them to generate an identity code by mapping the center of the cluster to a N-dimensional quantized bin. The resulting code is converted to a bit stream by a hashing mechanism to let user revoke his biometric in case of key compromise. This method presents the advantage of representing an individual by using his plenty of features instead of a single one in a supervised manner.