Advanced Search

Browse by Discipline

Scientific Societies

E-print Alerts

Add E-prints

E-print Network

  Advanced Search  

Reconstruction of source images from quantized biometric match score data

Summary: Reconstruction of source images from quantized
biometric match score data
Andy Adler
School of Information Technology and Engineering, University of Ottawa, Ontario, Canada
1. Introduction
One important consideration for the privacy and security of biometric systems is whether the enrolled images
may be regenerated from template or match score information. Many vendors of biometric systems have claimed
that it is impossible or infeasible to recreate the image from the templates; in light of this claim, biometric data has
been considered non-identifiable, and managed in ways that the source images cannot. For example, this assumed
non-identifiability has been used to allay concerns that fingerprint, face, and iris images may accessed from their
storage on identification cards.
Several authors have shown that it is possible to reconstruct a good estimate of an unknown enrolled image
from a fingerprint or face recognition template [1,5,8]. In each case, an algorithm is developed which allows small,
but physiologically realistic, modifications to be made to an input image. These modifications are used to perform a
"hill-climbing" attack. The modified image is presented to the algorithm and compared against the enrolled image to
obtain a match score. Modifications that increase the match score are retained. Eventually, a best-match image is
generated; this image resembles the essential features of the unknown enrolled image, and is able to compare to it at
high match score. The implication is that an attacker has effective access to the enrolled biometric image, which,
depending on the details of the security system, may have implications in terms of the system privacy and security.


Source: Adler, Andy - Department of Systems and Computer Engineering, Carleton University


Collections: Computer Technologies and Information Sciences