The authors analyse why Iris Exchange Report (IREX) VI conclusions about ‘iris ageing’ differ significantly from results of previous research on ‘iris template ageing’. They observe that IREX VI uses a definition of ‘iris ageing’ that is restricted to a subset of International Organization for Standardization (ISO)-definition template ageing. They also explain how IREX VI commits various methodological errors in obtaining what it calls its ‘best estimate of iris recognition ageing’. The OPS-XING dataset that IREX VI analyses for its ‘best estimate of iris recognition ageing’ contains no matches with Hamming distance >0.27. A ‘truncated regression’ technique should be used to analyse such a dataset, which IREX VI fails to do so, biasing its ‘best estimate’ to be lower-than-correct. IREX VI mixes Hamming distances from first, second and third attempts together in its regression, creating another source of bias towards a lower-than-correct value. In addition, the match scores in the OPS-XING dataset are generated from a ‘1-to-first’ matching strategy, meaning that they contain a small but unknown number of impostor matches, constituting another source of bias towards an artificially low value for ageing. Finally, IREX VI makes its ‘best estimate of iris recognition ageing’ by interpreting its regression model without taking into account the correlation among independent variables. This is another source of bias towards an artificially low value for ageing. Importantly, the IREX VI report does not acknowledge the existence of any of these sources of bias. They conclude with suggestions for a revised, improved IREX VI.

Bowyer and Ortiz, in their study ‘A Critical Examination of the IREX VI Results’, make seven criticisms of the authors application of linear mixed-effects models to longitudinally collected iris recognition Hamming distances. We reject these as either irrelevant, misinterpretations, or qualitatively correct, but quantitatively irrelevant.

The authors consider the problem of ‘re-identification’ where a biometric system answers the question ‘Has this person been encountered before?’ without actually deducing the person's identity. Such a system is vital in biometric surveillance applications and applicable to biometric de-duplication. In such a system, identifiers are created dynamically as and when the system encounters an input probe. Consequently, multiple probes of the same identity may be mistakenly assigned different identifiers, whereas probes from different identities may be mistakenly assigned the same identifier. In this study, they describe a re-identification system and develop terminology as well as mathematical expressions for prediction of matching errors. Furthermore, they demonstrate that the sequential order in which the probes are encountered by the system has a great impact on its matching performance. Experimental analysis based on unimodal and multimodal faces and fingerprint scores confirms the validity of the designed error prediction model, as well as demonstrates that traditional metrics for biometric recognition fail to accurately characterise the error dynamics of a re-identification system.

In this study, the authors will describe how performance results for continuous authentication (CA) should be reported. Most research on alleged CA is in fact periodic authentication, and performance is then reported in false match and false non-match rates. Here the authors will describe average number of impostor or genuine actions as the performance indicators, and will describe a more detailed performance reporting method. The authors’ current results have been reported in continuous authentication, based on analysis performed on two different datasets, and compared those results to the best results in comparable research, where they show that their results outperform most other known results.

Several researchers have presented studies of temporal effects on iris recognition accuracy, with varying results on severity of observed effects. The sensitive topic continues to be adversely discussed and the difficulty of isolating performance-impacting factors is immanent. The impact of ageing on segmentation vs. feature extraction has been largely neglected so far. This study attempts to shed light on the impact of segmentation quality on observed temporal effects highlighting the critical role of the segmentation module and quality assessment when assessing ageing effects. The lack of large and standardised temporal variation in public datasets as well as additional metadata (age of subject, recording parameters) and strictly enforced unified recording and quality guidelines over time-separated sessions are identified as imminent problems of ageing studies. Results are reported on a long-timespan database of 36,240 images comprising 104 classes and a 4 year time lapse, offering a large variety of recording conditions highlighting the critical role of a transparent recording setup.

The authors present an algorithm that models the rate of change of biometric performance over time on a subject-dependent basis. It is called ‘homomorphic users grouping algorithm’. Although the model is based on very simplistic assumptions that are inherent in linear regression, it has been applied successfully to estimate the performance of talking face and speech identity verification modalities, as well as their fusion, over a period of more than 600 days. Their experiments carried out on the MOBIO database show that subjects exhibit very different performance trends. Although the performance of some users degrades over time, which is consistent with the literature, they also found that for a similar proportion of users, their performance actually improves with use. The latter finding has never been reported in the literature. Hence, their findings suggest that the problem of biometric performance degradation may be not as serious as previously thought, and so far, the community has ignored the possibility of improved biometric performance over time. The findings also suggest that adaptive biometric systems, that is, systems that attempt to update biometric templates, should be subject-dependent.