Iris Aging Controversy In Biometrics
April Flowers for redOrbit.com – Your Universe Online
Will age affect the ability of biometric scanners to recognize your iris or not? This is an important question with an aging American workforce and the increasing reliance on technology for security.
Last year, redOrbit reported that the National Institute of Standards and Technology (NIST) released findings showing that no “consistent change occurs in the distinguishing texture of the iris for at least a decade.” A new study from the University of Notre Dame’s Kevin Bowyer and Estefan Ortiz points out errors of methodology in the NIST report and presents suggestions for revisions to be addressed in future versions of the “IREX VI: Temporal Stability of Iris Recognition Accuracy” report.
The debate rages on, according to Planet Biometrics, because the new NIST results are in direct conflict with several prior research studies from Notre Dame, Michigan State University, Clarkson University, West Virginia University and others. These previous studies analyzed different datasets, and all report observing significant iris template aging effect.
Bowyer, the Schubmehl-Prein Professor and the chair of the Department of Computer Science and Engineering, and Ortiz, a PhD candidate in the same department, point out various flaws in the IREX VI methodology, including using a definition of “iris aging” that is fundamentally different from the ISO standard definition of “iris template aging” used by previous researchers. This new definition is a sub-set of the overall definition. For example, previous studies have attempted to measure change in error rate for iris recognition over time, without excluding any possible cause for the increase in error rate. The IREX VI team, on the other hand, focused only on the change in error rate tied to appearance of the iris, and specifically excluded any changes in the iris related to pupil dilation. Bowyer and Ortiz concede that it is possible for the IREX report to be correct for the sub-set phenomenon that they studied, and the more general iris template aging to also be correct.
The more alarming part of the findings from Notre Dame includes claims of errors of methodology for the IREX report. Bowyer and Ortiz point out errors specifically in the regression analysis.
The first error is that the IREX researchers used a dataset that was “truncated” — meaning that all data points with a value over a certain threshold were deleted. The regression analysis performed on this truncated dataset did not take account of those data points, resulting in an estimate for iris aging that was biased to be lower than it should be in reality.
The second error Bowyer and Ortiz found concerns the regression analysis itself. The researchers assert that the IREX VI report takes one result out of context to represent the effects of iris aging. They say that the effects of aging may, in fact, be present in several results in the IREX VI regression model.
Bowyer and Ortiz point out a third methodology error: the data set used in IREX VI is a mixture of data points resulting from first, second and third attempts at iris recognition in the border-crossing application. This mixture could introduce bias that caused the estimated effect to be smaller than it is in reality.
The Notre Dame team included suggestions for improving a revised IREX VI report, including obtaining a new version of the data set used in the analysis, and using regression analysis methods appropriate to the data set.
Research findings from Notre Dame have played a role in previous IREX reports issued by NIST, including research on the effects of varying pupil dilation on the accuracy of iris recognition is discussed in the IREX I and the IREX III reports.