On the security of non-invertible fingerprint template transforms

Many transformation functions have been proposed for generating revocable or non-invertible biometric templates. However, their security analysis either ignores the distribution of biometric features or uses inefficient feature matching. This usually leads to unrealistic estimates of security. In this paper we introduce a new measure of non-invertibility, called the Coverage-Effort (CE) curve which measures the number of guesses (Effort) required by an adversary to recover a certain fraction (Coverage) of the original biometric data. In addition to utilizing the feature distribution, the CE curve allows estimation of security against partial recovery of biometric features. We analyze the CE curves obtained using different instances of a mixture of Gaussians based feature transform for fingerprint templates. Our analysis shows that knowledge of the fingerprint minutiae distribution reduces the effort required to obtain a specified coverage.