Can iris templates, which are used in a multitude of large-scale biometric systems around the world, age or not? It is an important question, and researchers just cannot seem to agree on the answers.
A significant, but controversial, test performed by NIST (called IREX VI) states plainly: “We find no evidence of a widespread iris ageing effect.”
The controversy arises because previous results by different research groups analyzing different datasets all report observing a significant iris template ageing effect.
So who is right?
No doubt the debate will continue to rage, and sensitivities will continue to run high. But a new report has been published by Kevin Bowyer and Estefan Ortiz of the Department of Computer Science and Engineering, University of Notre Dame, that attempts to explain why academic researchers have been achieving seemingly incompatible results.
The new report says that IREX VI uses a definition of “iris aging” that is restricted to only a subset of ISO-definition template aging. More alarmingly, the report also claims that in terms of regression analysis, IREX VI makes methodological errors in obtaining what it calls its “best estimate of iris recognition aging”.
According to the researchers: “The OPS - XING dataset that IREX VI analyzes for its ‘best estimate of iris recognition aging’ is truncated to remove all matches with Hamming distance greater than 0.27. To avoid bias in analyzing such a dataset, a “truncated regression” technique must be used. IREX VI fails to do so, and therefore its ‘best estimate of iris recognition aging” and other results are biased low.”
Other factors are also presented in the report that may have impacted NIST’s results.
According to the report, IREX VI does not attempt to study the same phenomenon studied in previous research, and so on that basis alone it cannot refute results of previous research.
Bowyer and Ortiz also say that the methodological errors mean that the IREX VI analysis should be re-done with appropriate techniques and controls to obtain unbiased results.
To read the full report click here.