Information theoretic capacity analysis for biometric hashing methods

In this paper, we address capacity analysis of biometric hashing methods. We propose an information theoretic capacity analysis framework for biometric hashing methods by taking into account their noise resilience which is analogous to the variations of the inputs for same user. To validate the proposed framework, we make simulations with various biometric hashing methods proposed in the literature on three different face image databases. We experimentally estimate the number of different users that a biometric hashing system can accommodate by assuming that every biometric hash vector is possible to be chosen for biometric template and within-class variations can be considered as noise and each bit position has the same probabilities. Besides, we calculate equal error rate performances of the biometric hashing methods and compare them with the proposed capacity analysis framework.