Assessing Fingerprint Individuality in Presence of Noisy Minutiae

Fingerprint image quality is an important source of intraclass variability. When the underlying image quality is poor, human experts as well as automatic systems are more likely to make errors in minutiae detection and matching by either missing true features or detecting spurious ones. As a consequence, fingerprint individuality estimates change depending on the quality of the underlying images. The goal of this paper is to quantitatively study the effect of noise in minutiae detection and localization, resulting from varying image quality, on fingerprint individuality. The measure of fingerprint individuality is modeled as a function of image quality via a random effects model and methodology for the estimation of unknown parameters is developed in a Bayesian framework. Empirical results on two databases, one in-house and another publicly available, demonstrate how the measure of fingerprint individuality increases as image quality becomes poor. The measure corresponding to the “12-point match” with 26 observed minutiae in the query and template fingerprints increases by several orders of magnitude when the fingerprint quality degrades from “best” to “poor.”