New study shows iris recognition is not infallible
New research has shown that iris recognition is not an infallible source for biometric data as earlier believed. Just like fingerprinting or facial recognition, iris recognition is susceptible to aging, thus enrolment for life is impossible.
These were the new findings by a University of Notre Dame research project conducted by Sam Fenker, an undergraduate student and Kevin Bowyer, Notre Dame’s Schubmehl-Prein Family Chair in Computer Science and Engineering.
The researchers went over a large dataset of images acquired over a longer period of time. From that, they were able to analyze iris changes over three successive years.
Bower noted: “The biometric community has long accepted that there is no ‘template aging effect’ for iris recognition, meaning that once you are enrolled in an iris recognition system, your chances of experiencing a false non-match error remain constant over time. This was sometimes expressed as ‘a single enrollment for life.’ Our experimental results show that, in fact, the false non-match rate increases over time, which means that the single enrollment for life idea is wrong.”
There are two ways of checking a false match rate. One determines how often the system says ‘match’ on the two images coming from different persons. The second determines how often the system says ‘no match’ on two images coming from the same person.
Bowyer said, the result should not be perceived as a “negative” for iris recognition technology but rather a step towards finding new methods to deal with aging.
“Once you have admitted that there is a template effect and have set up your system to handle it appropriately in some way, it is no longer a big deal,” he said. “One possibility is setting up a re-enrollment interval. Another possibility is some type of ‘rolling re-enrollment,’ in which a person is automatically re-enrolled each time they are recognized. And, in the long run, researchers may develop new approaches that are ‘aging-resistant.’ The iris template aging effect will only be a problem for those who for some reason refuse to believe that it exists.”
What do you think is more stable, iris recognition or fingerprinting as sources of biometrics?