Iris patterns are generally thought of as being unique and stable throughout a person's life - indeed these characteristics are crucial in order to make any good biometric trait. New research, however, has been reported that questions just how stable an iris is over time, something that could affect the normally impressive performance of iris recognition systems on large populations.
An article in New Scientist reports that iris scans from a particular eye can alter over time, potentially meaning that a system will fail to match an iris if the template it matches against is very old.
The structures creating distinctive iris patterns are complete by the eighth month of gestation (although pigmentation continues into the first years after birth). However, Kevin Bowyer at the University of Notre Dame in Indiana and his colleagues say they now have evidence that the Hamming distance increases over time for scans of the same eye. Hamming distance is an important statistical measure - essentially if the Hamming distance is found to be below the decision threshold, then a positive identification is made. So if Hamming distance increases over time then there is an increased chance of false rejection by the system.
Bowyer and Co. initially compared pairs of scans of 26 irises taken two months apart with scans made four years apart. According to the report in New Scientist (based on recently submitted analysis) the team measured an increase of 0.018 per cent over the period.
So does this make a difference? Especially given the scale of some projects now under consideration, such as the UIDAI scheme in India. Probably not, it seems…
"Even if this finding is true, it is no cause for concern," says John Daugman of the University of Cambridge. He says a real-world system may have a false rejection rate of the order of 1 in 250 million, so the increase in Hamming distance measured by Bowyer would result in a false rejection rate of just 0.7 in 100 million scans after four years.