A new report by biometric researchers at the National Institute of Standards and Technology (NIST)used data from thousands of frequent travelers enrolled in an iris recognition program to determine that no consistent change occurs in the distinguishing texture of their irises for at least a decade.
The new study counters a previous study of 217 subjects over a three-year period that found that recognition of the subjects’ irises became increasingly difficult, consistent with an aging effect.*
To learn more, NIST biometric researchers used several methods to evaluate iris stability.
A frequent traveler uses an iris recognition camera to speed her travel across the American-Canadian border. NIST researchers evaluated data from millions of images taken over a decade from this iris-based NEXUS program to gauge iris stability. (Credit: Canadian Border Services Agency)
Researchers first examined anonymous data from millions of transactions from NEXUS, a joint Canadian and American program used by frequent travelers to move quickly across the Canadian border.
As part of NEXUS, members’ irises are enrolled into the system with an iris camera and their irises are scanned and matched to system files when they travel across the border.
NIST researchers also examined a larger, but less well-controlled set of anonymous statistics collected over a six-year period.
No evidence of a widespread aging effect
In both large-population studies, NIST researchers found no evidence of a widespread aging effect, said Biometric Testing Project Leader Patrick Grother. A NIST computer model estimates that iris recognition of average people will typically be useable for decades after the initial enrollment.
“In our iris aging study we used a mixed effects regression model, for its ability to capture population-wide aging and individual-specific aging, and to estimate the aging rate over decades,” said Grother. “We hope these methods will be applicable to other biometric aging studies such as face aging because of their ability to represent variation across individuals who appear in a biometric system irregularly.”
NIST researchers then reanalyzed the images from the earlier studies of 217 subjects that evaluated the population-wide aspect. Those studies reported an increase in false rejection rates over time—that is, the original, enrolled images taken in the first year of the study did not match those taken later.
While the rejection numbers were high, the results did not necessarily demonstrate that the iris texture itself was changing. In fact, a study by another research team identified pupil dilation as the primary cause behind the false rejection rates.**
NIST researchers showed that dilation in the original pool of subjects increased in the second year of the test and decreased the next, but was not able to determine why. When they accounted for the dilation effect, researchers did not observe a change in the texture or aging effect. Some iris cameras normalize dilation by using shielding or by varying the illumination.
NIST established the Iris Exchange (IREX) program in 2008 to give quantitative support to iris recognition standardization, development and deployment. Sponsors for this research include the Criminal Justice Information Systems Division of the Federal Bureau of Investigation, the Office of Biometric Identity Management in the Department of Homeland Security (DHS) and the DHS Science and Technology Directorate.
The findings do not address privacy issues. This coming fall, numerous schools from every different level, college to elementary, will be incorporating eyeball scans instead of IDs. A Fox News report (below) expresses concern with privacy and loss of parental control in schools in Florida, while a Geek Insider writer thinks it’s a great idea.
What do you think?
*S. Fenker and K.W. Bowyer. Experimental evidence of a template aging effect in iris biometrics. IEEE Computer Society Workshop on Applications of Computer Vision, November 2012.
**M. Fairhurst and M. Erbilek. Analysis of physical ageing effects in iris biometrics. IET Computer Vision, 5(6):358–366, 2011. ww.ietdl.org.