Iris aging - Pitfalls in studying “Big Data” from operational scenarios
10 August 2016 17:39 GMT

If you want to glean better - more valid - results in a given scientific study (in this case iris biometrics) it surely makes sense to extract your findings from as large an "operational" dataset as possible. Right? Not necessarily, say researchers from the University of Notre Dame's Department of Computer Science and Engineering.

In a new paper to be published later this year (at BTAS 2016), Estefan Ortiz and Kevin Bowyer, argue that while it may seem like common sense to study as large an operational biometric dataset as possible, so as to "guarantee" valid results, a number of basic questions should be asked in order to avoid the possibility that the data is not suitable.  

The researchers explored their hypothesis using a large operational iris recognition dataset from the Canada Border Services Agency’s NEXUS program, similar to the dataset analyzed in the NIST IREX VI report.

At the core of the paper is an ongoing disagreement around the possible effect of iris aging, which was dismissed as being significant by the IREX VI report, but noted in smaller studies by Bowyer and colleagues.

The paper describes how big data, and especially big data from an operational scenario, can contain subtle and unanticipated complexities. It points out how these complexities can complicate analysis of such datasets to answer research questions. In particular, complexities that arise in a large, operational dataset can make it difficult to obtain a meaningful answer to an apparently simple research question such as iris template aging, say the researchers. 

Factors at play include the geographic location the iris data is collected, seasonal variations, and lighting.

So. Can the iris template aging mystery now be solved? It appears not. The researchers say that given the number and variety of uncontrolled factors involved in the data collection, it would be speculative at this point to make any conclusion about “the” cause of any time-varying trend in the match scores.

In future research, they hope to develop a better understanding of the seasonal and location-dependent fluctuations in the iris match scores.

Related articles

Journal to explore iris ageing debate
No evidence of widespread iris ageing effect, says NIST