Received February 26, 2013, accepted May 3, 2013, date of publication May 16, 2013, date of current version

May 21, 2013.

Digital Object Identifier 10.1109/ACCESS.2013.2262916

Template Aging Phenomenon in Iris Recognition

SAMUEL P. FENKER, ESTEFAN ORTIZ, AND KEVIN W. BOWYER (Fellow, IEEE)
Department of Computer Science and Engineering, University of Notre Dame, Notre Dame, IN 46556, USA

Corresponding author: K. W. Bowyer (kwb@cse.nd.edu)

ABSTRACT Biometric template aging is dened as an increase in recognition error rate with increased time since enrollment. It is believed that template aging does not occur for iris recognition. Several research groups, however, have recently reported experimental results showing that iris template aging does occur. This template aging effect manifests as a shift in the authentic distribution, resulting in an increased false nonmatch rate. Analyzing results from a three-year time-lapse data set, we nd 150% increase in the false non-match rate at a decision threshold representing a one in two million false match rate. We summarize several known elements of eye aging that could contribute to template aging, including age-related change in pupil dilation. Finally, we discuss various steps that can control the template aging effect in typical identity verication applications. INDEX TERMS Biometrics, iris recognition, error probability, false non-match rate, template aging.

I. INTRODUCTION

Iris recognition technology began in the early 1990s with the work of John Daugman [1], [2], and has progressed quickly. The United Arab Emirates has used iris recognition successfully in border control for the past decade [3]. India is using iris recognition as part of its Aadhaar, or Unique ID, program to give a unique ID number to each of about 1.2 billion citizens. In 2013, the number of persons enrolled in Indias Aadhaar program will surpass the size of the population of the United States [4]. In addition to these large government applications, there are successful uses of iris recognition in banks, hospitals, schools and other applications. Along with the expanding range of applications for iris recognition, or perhaps because of it, iris recognition is also an active and expanding research area [5], [6]. This paper is concerned with template aging in iris recognition. The ISO standard for biometric performance testing denes template aging as follows Longer time intervals generally make it more difcult to match samples to templates due to the phenomenon known as template aging. This refers to the increase in error rates caused by timerelated changes in the biometric pattern, its presentation and the sensor [7]. Template aging receives substantial research attention in the face recognition research community (e.g., [8], [9]) and is documented in ngerprint matching [10]. However, template aging has, until recently, been assumed to not occur for iris recognition. Section II reviews the literature related to template aging in iris recognition. Section III details our own recent
266

experimental results on iris template aging. Section IV outlines how various elements of normal eye aging could contribute to an iris template aging effect. Section V describes non-traditional enrollment schemes that could be implemented to reduce or eliminate the template aging effect. Section VI summarizes some of the issues arising in connection with research on iris template aging.
II. LITERATURE REVIEW

In their 1987 patent, Flom and Sar [11] asserted that, . . . the signicant features of the iris remain extremely stable and do not change over a period of many years. However, they also acknowledged that the iris does change over time, and that reenrollment might be needed [11] Even features which do develop over time . . . usually develop rather slowly, so that an updated iris image will permit ID for a substantial period . . . . Daugmans 1994 patent [1] asserted that the iris texture is unchanging for life The iris of every human eye has a unique texture of high complexity, which proves to be essentially immutable over a persons life. Elsewhere, Daugman indicates that this conclusion was supported through subjective evaluation of a dataset of ophthalmologists images [2] The clinical database of iris images made available to this author from ophthalmologists photographs spanning a 25 year period did not reveal any noticeable changes in iris patterns for individual subjects. The view that iris texture is essentially immutable over a persons life became the prevailing view in the iris recognition community. The assumption is often repeated in
VOLUME 1, 2013

2169-3536/$31.00
2013 IEEE

S. P. FENKER et al.: Template Aging Phenomenon

the research literature, e.g., . . . the iris is highly stable over a persons lifetime . . . [12], [the iris is] essentially stable over a lifetime [13], and the iris is highly stable over a persons lifetime [14]. Popular references such as Wikipedia often expressed it as a key advantage . . . is . . . template longevity, as, barring trauma, a single enrollment can last a lifetime [15]. The idea that a single enrollment can last a lifetime is challenged or disproved by recent research on iris template aging. Tome-Gonzalez et al [16] compare matches between images acquired in the same session with matches between images acquired with one to four weeks of time lapse. They use two different datasets, each acquired across four weekly sessions, using an LG 3000 sensor. They use Maseks matcher [17], which does not have state-of-the-art performance. At a false match rate (FMR) of 0.01%, they report a false nonmatch rate (FNMR) of 8.5% to 11.3% for within-session matches, versus a FNMR of 22.4% to 25.8% for acrosssession matches. This is a same-session versus across-session comparison, rather than a longitudinal template aging study, but it shows that iris match quality can depend on factors that change between acquisition sessions. Baker et al [18] report on a study involving 26 irises (13 persons) with images acquired over 20042008 using an LG 2200 sensor. They used a version of the IrisBEE matcher that was distributed in the Iris Challenge Evaluation [19]. This matcher does not have state-of-the-art performance. They compared the authentic and impostor distributions for short-term and long-term matches. Short-term matches were between two images taken in the same academic semester but not on the same day, in order to avoid same-session bias. Long-term matches were between an image taken in spring of 2004 and one taken in spring of 2008. They found that the authentic distribution for long-term matches shifted so as to increase the FNMR. They reported that, at a false accept rate of 0.01%, the false reject rate increases by 75% for long-timelapse [18]. They later expanded this study to include a larger dataset and additional matchers [46]. Rankin et al [20] studied variation in iris appearance over three sessions at three-month intervals. They used visiblelight illumination rather than near-infrared. We are not aware of any commercial iris recognition system that uses visible light. The quality of their images and iris matching software was such that Recognition failure was detected in 21% of intra-class comparisons cases overall, taken at both three and six month intervals [20]. The combination of short time lapse, use of visible-light images, and such poor overall recognition accuracy makes it impossible to draw any reliable conclusions related to iris template aging. Sazonova et al [21] studied iris template aging using a dataset acquired at Clarkson University with an Oki IrisPass-H sensor over the period 20052007 for 46 subjects. They analyzed results from the Masek [17] and VeriEye [22] matchers. They also looked at results based on using image quality metrics to select a subset of good quality images. They found a template aging effect regardless of matcher
VOLUME 1, 2013

or image quality subset All verication rates decrease with the increase of the time lapse [21]. Results for the good quality subset of the overall image dataset showed a more pronounced template aging effect (see Table 1 of [21]), which suggests that while quality varies over time, it may be masking the effect rather than contributing to the effect [21]. In our own earlier work, using data for 86 irises (43 persons), we compared short-term matches, between two images taken on different days within one semester, with long-term matches, between images taken in spring 2008 and spring 2010 using an LG 4000 sensor. We found that the FNMR increased for long-term matches. For the IrisBEE matcher, we found that the increase in false reject rate ranges from 157% at a threshold of 0.28 to 305% at 0.34 [23]. For the VeriEye matcher [22], we found that The observed false reject rate increases from short to long time-lapse by 195% at a threshold of 30 and up to 457% at a threshold of 100 [23]. Later results [24] expanded on this study to include a third year of time lapse, by presenting results for one, two and three years, and by using a bootstrap method [25] to compute 95% condence intervals for the estimated change in the FNMR. This current paper builds on the results in [24] by explicating several possible contributing factors in iris template aging, and also by discussing several possible means to control the effects of iris template aging. Czajka [26] studied iris template aging with a dataset of 571 images representing 58 eyes, with up to eight years time lapse, acquired between 2003 and 2011. He considered performance of three different iris matchers, and concluded that [26] Average values of the genuine scores obtained for all the tested matchers may suggest that iris templates age, what partially supports earlier ndings determined for different matchers and different databases, yet collecting samples with a shorter time lapse between captures than in this work. He also reports that [26] The extent to which the template ageing phenomenon is observed is however uneven across different matchers, in particular we observe a higher inuence of the time ow for more accurate methods. Ellavarason and Rathgeb [44] re-analyzed the two-year time-lapse dataset of Fenker and Bowyer [23] using six different iris feature extraction algorithms. The same weighted adaptive Hough and ellipsopolar transform [45] is used for iris segmentation for each of the six iris feature extraction algorithms. The highest-performing of the six algorithms is generally the 1-D log-Gabor ltering based on Maseks work [17]. In general, for the six algorithms, they report clear evidence of template aging . . . the fact that the resulting graph shows a gap between the FNMR rates for short and long term shows the existence of ageing [44]. In summary, recent studies by different research groups have found an iris template aging effect. These studies involve various different image datasets and various matching algorithms, various approaches to analysis, and explore lengths of time from two to eight years. All report an increase in the FNMR with increased time since enrollment. There are
267

S. P. FENKER et al.: Template Aging Phenomenon

also suggestions that template aging may be more pronounced for higher quality images [21] and for more accurate matchers [26].
III. IRIS RECOGNITION TEMPLATE AGING EXPERIMENT

TABLE 1. Experimental data by period of time lapse

Time Period 08-09 09-10 10-11 08-10 09-11 08-11

Number of Subjects 88 157 181 40 124 32

Number of Images 4,553 8,046 11,734 2097 8,082 2,338

Iris images for this experiment were acquired from 2008 through 2011 using an LG 4000 sensor. Example images of the same iris from 2008 and 2011 are shown in Figure 1. Of the 322 total subjects, 177 are male and 145 are female; and 243 are Caucasian, 37 are Asian, 24 are other and 18 unknown. Subject age varies from 20 to 64 years old. We consider the dataset in terms of one cohort with three years of time lapse (20082011), two cohorts with two years (20082010 and 20092011), and three cohorts with one year (20082009, 20092010 and 20102011). Table 1 summarizes the numbers of subjects, images, and short- and long-term authentic matches.
A. SELECTION OF IRIS MATCHER

Short TimeLapse Matches 11,986 23,882 29,120 5,829 18,963 5,244

Long TimeLapse Matches 30,470 54,417 97,879 14,282 66,849 20,888

We evaluated four iris matchers available to us in order to select the best-performing one for use in this experiment. One is our own version of the IrisBEE matcher. The second is the commercial VeriEye SDK (version 2.4) [22]. The third and fourth matchers are other commercial SDKs,

which are not named here due to restrictions in their license agreements. Using the entire image dataset described above, an allversus-all matching experiment was performed with the four matchers. The results, illustrated in Figure 2, involve over 285 million comparisons. Based on the VeriEye matcher performing the best of the four matchers in this experiment, it was selected for use in the remaining experiments. It is worth noting that VeriEye was also found to be one of the topperforming matchers in the recent NIST IREX report [41]. Unlike a matcher that reports a fractional Hamming distance, the VeriEye matcher reports a similarity score that ranges from 0 to 9443, where 0 is a non-match and 9443 is an exact match; that is, an image matched against itself.
B. EXPERIMENTAL METHOD

We dene a short-time-lapse match to be a comparison of two images acquired within a few months of each other, but on different days, so that there are no same session matches. For our experiments, we have a set of short-timelapse matches for images acquired in the spring semester of each of 2008, 2009 and 2010. We dene a long-time-lapse match as a comparison of two images acquired in different years. For example, we have a set of long-time-lapse matches between an image from spring 2008 and an image from spring 2009. We have three different one-year time-lapse datasets, two different two-year datasets, and one three-year dataset. The short-time-lapse spring 2008 dataset is the baseline for the one-year 20082009 dataset, for the two-year 20082010 dataset, and the three-year

0.995

True Accept Rate

0.99 IrisBee 0.985 VeriEye Commercial 1 0.98 Commercial 2

0.975

0.97 0 0.005 0.01

False Accept Rate

0.015

0.02

0.025

0.03

FIGURE 1. Images of the Same Eye from 2008 (top) and 2011 (bottom). Images were acquired using an LG 4000 iris sensor. 268

FIGURE 2. Selection of a best-performing matcher to use in template-aging experiments.

VOLUME 1, 2013

S. P. FENKER et al.: Template Aging Phenomenon

20082011 dataset. The short-time-lapse 2009 dataset is the baseline for the one-year 20092010 dataset and the two-year 20092011 dataset. The short-time-lapse spring 2010 dataset is the baseline for the one-year 20102011 dataset. We compared the impostor and authentic distributions for each short-time-lapse baseline to those of each of its longtime-lapse comparisons. We found that the impostor distribution for matches between images taken on average about one month apart does not differ substantially from the impostor distribution for one, two or three years of time lapse. However, the authentic distribution for a long time-lapse dataset has a generally higher FNMR than its corresponding shorttime-lapse distribution. To investigate this further, we compared the FNMR for short and long time lapse, across a range of decision threshold values. We also calculated a bootstrap 95% condence interval [25] for the estimated change in FNMR. Statistics are noted for the various datasets for a particular decision threshold value based on a false match rate (FMR) selected based on the experiment shown in Figure 2. For the VeriEye matcher, in general, at a low enough value for the decision threshold, the FMR is 100% and the FNMR is 0%. As the decision threshold is increased, the FMR decreases and the FNMR increases. For the experiment shown in Figure 2, a decision threshold of 580 corresponds to an approximately 1 in 2 million FMR. At this threshold, there is sufcient FNM data in the various datasets to justify computing bootstrap estimates of the change in the FNMR. At the decision threshold corresponding to a 1 in 1 million FMR, there generally are too few FNM results to allow a reliable estimation of the FNMR.
C. ONE-YEAR TIME-LAPSE RESULTS

FIGURE 3. Comparison of Short- and Long-Term FNMR for 20082009.

This corresponds to a mean percent increase in FNMR of 27%, with a condence interval of 5% to 61%. The FNMR as a function of the decision threshold for the 20092010 dataset is qualitatively the same as that shown in Figure 3 for 20082009. Again, the FNMR curves separate as soon as the FNMR is measurable and the gap between them increases as the decision threshold increases. At the 1-in-2MFMR threshold, the mean absolute increase in the FNMR is 0.086, with 95% condence interval of 0.063 to 0.112. The corresponding percent mean increase in the FNMR is 60%, with a 95% condence interval of 40% to 84%. The FNMR as a function of decision threshold for the 20102011 dataset is also qualitatively the same. Again, the FNMR curve for long time-lapse consistently runs above that for short time-lapse. At the 1-in-2M-FMR threshold, the mean absolute increase in FNMR is 0.063, with 95% condence interval of 0.041 to 0.086. The corresponding percent increase in the FNMR is 49%, with a 95% condence interval of 29% to 73%.
D. TWO-YEAR TIME-LAPSE RESULTS

The three one-year time-lapse datasets represent 226 irises (113 subjects) imaged in 20082009, 338 irises (169 subjects) imaged in 20092010, and 406 irises (203 subjects) imaged in 20102011. The average number of days between acquiring a pair of images in the short-time-lapse match group is approximately 39 days, and the average number of days between acquiring a pair of images in the long-time-lapse match group is approximately 360 days. Thus the average aging between the two authentic distributions is approximately 11 months. The FNMR as a function of the decision threshold for the 20082009 dataset is shown in Figure 3. The FNMR starts at zero for a low decision threshold value and increases as the decision threshold increases. If no template aging effect existed, then the FNMR curves for the short and long time-lapse datasets should be the same. Figure 3 shows that the FNMR curves for short and long time-lapse separate as soon as the FNMR is measurable and that the gap between the curves increases over the range of decision threshold values. At the decision threshold of 580, corresponding to approximately a 1 in 2 million FMR on the overall dataset, the bootstrap mean increase in the absolute FNMR is 0.035, with a 95% condence interval of 0.00723 to 0.0627.
VOLUME 1, 2013

We have two datasets representing two-year time lapse. These represent 90 irises (45 persons) in the 20082010 dataset and 270 irises (135 persons) in the 20092011 dataset. The average number of days between a pair of images in the twoyear time-lapse data is 732 days. The FNMR as a function of the decision threshold for the two-year, 20082010 dataset is shown in Figure 4. As is the case with the three one-year datasets, the FNMR curve for long time-lapse runs above that for short time-lapse. At the decision threshold corresponding to a 1-in-2M FMR, the mean absolute increase in the FNMR is 0.11, with a 95% condence interval of 0.056 and 0.168. This corresponds to a mean percent increase in the FNMR of 82%, with a 95% condence interval of 38% to 150%. For the two-year, 20092011 dataset, the FNMR for long time-lapse is again consistently greater than that for short time-lapse. At the decision threshold for a 1-in-2M FMR, the mean absolute increase in the FNMR is 0.13, with a 95% condence interval of 0.095 to 0.167. This corresponds to a mean percent increase in the FNMR of 91%, with a 95% condence interval of 63% to 127%.
269

S. P. FENKER et al.: Template Aging Phenomenon

FIGURE 4. Comparison of Short- and Long-Term FNMR for 20082010.

FIGURE 5. Comparison of Short- and Long-Term FNMR for 20082011.

For each of the two two-year time-lapse datasets, as was the case with each of the three one-year time-lapse results, there is a clear template aging effect. At the 1-in-2M FMR decision threshold, the mean FNMR for the two datasets is estimated at 82% and 91%. Each of these two values is greater than the value for any of the three one-year datasets. And the lower limit of the 95% condence interval does not include zero change in the FNMR.
E. THREE-YEAR TIME LAPSE (20082011)

FNMR with increasing time lapse. This is consistent evidence of an iris template aging effect. However, it is also important to note that the 95% condence interval indicates a broad range for the estimate of the increase in the FNMR.
IV. EYE AGING AND IRIS TEMPLATE AGING

There are 70 irises (35 subjects) in the 20082011 time-lapse dataset. The mean time between image acquisitions for the two images in a long-time-lapse match is 1,068 days. Thus the average time lapse is about 34 months. The FNMR as a function of the decision threshold for the three-year, 20082011 dataset is shown in Figure 5. Again, the FNMR for long time-lapse is consistently greater than for short time-lapse. At the 1-in-2M FMR decision threshold, the mean absolute increase in the FNMR is 0.147, with a 95% condence interval of 0.088 and 0.208. This corresponds to a mean percent increase in the FNMR of 153%, with a 95% condence interval 85% to 307%. This mean percent increase for three-year time lapse is greater than that for either of the two-year datasets.
F. CHANGE BETWEEN 20082009, 20082010 AND 20082011

The one-year 20082009 results, the two-year 20082010 results, and the three-year 20082011 results share highly overlapping short time-lapse authentic distributions based on images from spring 2008. The long time-lapse authentic distributions represent the spring 2008 subjects who also had iris images acquired in spring 2009, 2010 and 2011, respectively. Consider these three results as a time sequence. For 20082009, the mean percent increase in FNMR is 27%, with a condence interval of 5% to 61%. For 20082010, the mean percent increase in the FNMR is 82%, with a 95% condence interval 38% to 150%. For 20082011, the mean percent increase in the FNMR is 153%, with 95% condence interval for the percent increase is 85% to 307%. This data, summarized in Table 2, shows a clear trend of increasing
270

The biometric signal for iris recognition is the texture of the iris. Of course, the features for iris recognition are computed from an image of the iris texture rather than the iris itself. This distinction is important because any eye aging effect that changes the image of the iris texture automatically has the potential to cause template aging. Consider the structure of the eye as depicted in Figure 6. The iris is necessarily imaged through the cornea and the aqueous humor, and subject to possible occlusion in the image by the eyelids and eyelashes. The human eye is known to undergo many different agerelated changes [27]. We briey outline how known agingrelated effects involving the eyelids, the cornea, and the iris could contribute to template aging. This list is not at all intended to be comprehensive. It is only to suggest the complexity of a comprehensive consideration of possible factors contributing to template aging. Establishing a comprehensive list of eye aging mechanisms that can contribute to iris template aging, and the relative importance of their contributions, would be a substantial research topic in itself. The eyelids can occlude part of the iris in an image. The term ptosis refers to drooping of the eyelid. It is known that the drooping of the eyelid can be caused by the normal aging process [28]. Eyelid droop that increases with age can result in increased iris occlusion in later images. This translates into less of the iris in view in common between two images as there is more elapsed time between the images. Less iris area in view in common between the images effectively means fewer iris code bits for matching, which in turn means a larger variance for the match score and an increased probability of error. Thus ptsosis, or eyelid droop, can potentially contribute to an iris template aging effect. The cornea is often idealized as having spherical-shaped outer (anterior) and inner (posterior) surface. Models of iris image formation that use such a 3D model of the cornea have recently been explored as a means to correct offangle iris images [29][31]. As Kennell et al [30] point out,
VOLUME 1, 2013

S. P. FENKER et al.: Template Aging Phenomenon

TABLE 2. FNMR from 20082009 to 20082011

Time Period 20082009 20082010 20082011

% Change in FNMR (95% CI) 27% (5%, 61%) 82% (38%, 150%) 153% (85%, 307%)

FIGURE 7. Sequence of Iris Images Ordered By Pupil Dilation Ratio. This video is formed from a sequence of iris images taken at different times and sequenced in order of increasing pupil dilation ratio. It is clear that there are changes in iris texture based on pupil dilation, and that larger differences in texture occur for larger differences in dilation (please see multimedia package to view video).

FIGURE 6. Diagram of the Structure of the Eye. (Courtesy of National Eye Institute, National Institutes of Health; Ref#: NEA05).

The shapes and measurements of the cornea and iris will vary from person to person, and indeed the topography of the cornea is irregular, and thus the cornea cannot be perfectly modeled as a smooth spherical or ellipsoidal surface. Not mentioned in [30], but important in the context of this paper, is that the shape of the cornea changes with age [32], [33]. At younger ages, the cornea tends to have greater curvature along the horizontal axis than along the vertical axis. At older ages, the opposite is true. This is referred to as a change from with the rule to against the rule astigmatism. In a study involving over 700 persons from aged 20 to 80-plus, Hayashi et al [32] note, the normal cornea becomes steeper and shifts from with-the-rule to against-the-rule astigmatism with age. In a similar study, Vihlen and Wilson found [33] a signicant relationship between corneal toricity and age, i.e., corneal toricity shifts towards against-the-rule with age. The distance from the corneal surface to the iris also changes with age. The anterior chamber is the part of the eye from the outer surface of the cornea to the iris. Atchison et al [34] reported that for the emmetropic eye (an eye not needing vision correction), Despite considerable data scatter, we found signicant age changes: anterior chamber depth decreased 0.011 mm/year . . . . Changes in the shape of the cornea and in the distance from the cornea surface to the iris surface will cause images of the same iris taken at different ages to have differently
VOLUME 1, 2013

warped versions of the iris texture. Thus, age-related changes in cornea shape are a potential contributor to template aging. The iris itself undergoes known functional changes with age [35][37]. In a study of pupil size across persons of different ages and at different illumination levels, Winn et al [36] concluded pupil size decreased linearly as a function of age at all illuminance levels. (In addition to the average pupil size becoming smaller with age, the responsiveness of the iris to a given change in light intensity decreases with age [37].) Because average pupil size decreases with age, increasing time between images of the same iris results in increasing average difference in pupil dilation. Studies in the iris recognition literature show that increased difference in pupil dilation results in an increased FNMR [38], [39]. See Figure 7 for an illustration of how iris texture changes with pupil dilation. Thus the normal aging effect of a decreasing average pupil size with increasing age is a contributor to iris template aging. Figure 8 shows a bar chart of the mean difference in pupil dilation for the datasets whose pattern of increase in FNMR is summarized in Table 2. The difference in pupil dilation for a pair of images is computed as the absolute value of the difference in the PupilIrisRatio value given by the VeriEye SDK. Note that the average difference in pupil dilation ratio increases with increased time, just as is to be expected based on what is known about iris aging [35], [36]. The increased difference in pupil dilation is then expected, based on the iris recognition literature [38], [39], to degrade the match scores. The contribution of increased difference in pupil dilation to template aging has been noted in previous work [23].
271

S. P. FENKER et al.: Template Aging Phenomenon

FIGURE 8. Difference in Pupil Dilation for 20082008, 20082009, 20082010 and 20082011. The 20082008 value is the mean difference in dilation ratio between two images of the same eye acquired in 2008, in units of the VeriEye SDKs PupilIrisRatio value, which has a range of 0 to 255; similarly with 20082009, 20082010 and 20082011.

While pupil dilation is not the only factor contributing to template aging, it appears to be an important factor. It is also currently the most well-explored factor in template aging. It is possible in principle to study the impact of eyelid occlusion from the image data. However, the appropriate parameters are not directly reported in the VeriEye SDK used in this work. Investigation of factors such as change in corneal shape does not appear to be possible using only the type of images acquired for iris recognition. It is sometimes asserted that the iris texture does not change with age and that therefore template aging does not occur. Setting aside the question of whether the iris texture changes with age, there is something important to note about the aging effects outlined above. Even if the iris texture at a given dilation ratio was perfectly constant throughout a persons lifetime, each of the mechanisms outlined above could still contribute to template aging. This demonstrates how assertions about the apparent constancy of iris texture over time can simply miss the point when it comes to discussions about template aging.
V. APPROACHES TO CONTROLLING TEMPLATE AGING

This makes it possible to limit the template aging effect due to difference in pupil dilation by matching against multiple templates representing varying degrees of pupil dilation. A second step to limit the performance degradation due to various other sources of template aging is to automatically update an enrollment template based on a new iris image used for a successful verication. The template that is updated should be the one that represents a dilation ratio closest to that of the new image. It may be that a more stringent set of quality checks is required for a verication that generates a template update and that a less stringent set of checks allows verication without a template update. The general effect of this approach is to limit the aging period in a practical way. The general family of schemes in which a successful verication can result in an updated enrollment template has been termed rolling re-enrollment. It is also possible that the iris sensor and / or the image acquisition environment could use visible-light illumination to attempt to control pupil dilation. The LG 4000 sensor used to acquire images for our study described above does not use visible-light illumination to attempt to control pupil dilation. However, the Iris Guard AD 100 sensor [42] appears to monitor pupil dilation during acquisition and if the pupil dilation is too large, to switch on a visible light to induce a smaller pupil size. Thus the control of template aging due to difference in pupil dilation can have both algorithm-focused and sensor-focused elements.
VI. SUMMARY AND DISCUSSION

While it was once believed that iris template aging did not occur and so a single enrollment can last a lifetime, multiple studies now show that iris template aging does occur [18], [21], [23], [24], [26]. Specically, the experimental evidence shows that the FNMR increases with increasing time between the enrollment image and the image to be recognized. The template aging effect found in these studies can be controlled by relatively simple enhancements of the traditional methods of enrolling and matching irises. Several points seem important to make for a better understanding of the iris template aging issue.
1) NORMAL AGE-RELATED CHANGE IN PUPIL DILATION IS ONE CONTRIBUTOR TO IRIS TEMPLATE AGING

Historically, the standard practice for iris enrollment is to enroll a template computed from a single image that passes checks for focus quality and percent of iris occluded. The standard practice for matching is to match a template from a single image. Given what is currently known about iris template aging, it appears that a combination of relatively simple extensions to the normal iris enrollment and matching procedure could effectively control iris template aging for many identity verication applications. One step to effectively eliminate the component of template aging due to average pupil size decreasing with age is to enroll an iris with a set of templates from multiple images that represent a range of dilation values. The initial enrollment process might then have a more active element of varying illumination in order to generate a range of pupil dilation, with special attention to obtaining images with smaller pupil sizes.
272

It is known that average pupil dilation decreases with age [35], [36]. It is also known that increased difference in pupil dilation increases the FNMR [38], [39]. Thus agerelated change in pupil dilation is one contributing cause to iris template aging. Other possible contributing factors in iris template aging remain to be studied in detail.
2) RELATIVELY SIMPLE CHANGES TO ENROLLMENT AND MATCHING CAN CONTROL TEMPLATE AGING

Iris template aging is made more pronounced by the traditional practice of enrolling a single template and assuming that it is for life. Enrolling an iris with multiple templates representing different degrees of dilation improves iris recogVOLUME 1, 2013

S. P. FENKER et al.: Template Aging Phenomenon

nition accuracy [43]. In situations where the biometric is used for access control, a rolling re-enrollment1 scenario may be an attractive solution. A users set of enrollment templates may be automatically updated after a verication in which the image passes appropriate quality control checks.
3) TEMPLATE AGING CAN OCCUR EVEN IF THE IRIS TEXTURE REMAINS THE SAME

REFERENCES
[1] J. G. Daugman, Biometric personal identication based on iris analysis, U.S. Patent 5 291 560, Mar. 1, 1994. [2] J. Daugman, Recognizing persons by their iris patterns, in Biometrics: Personal Identication in Networked Society, A. K. Jain, R. Bolle, and S. Pankanti, Eds. Norwell, MA, USA: Kluwer, 1998. [3] A. N. Al-Raisi and A. M. Al-Khouri, Iris recognition and the challenge of homeland and border control security in UAE, Telemat. Informat., vol. 25, no. 2, pp. 117132, 2008. [4] Unique Identication Authority of India (2013, May 15) [Online]. Available: http://uidai.gov.in [5] K. W. Bowyer, K. Hollingsworth, and P. J. Flynn, Image understanding for iris biometrics: A survey, Comput. Vis. Image Understand., vol. 110, no. 2, pp. 281307, May 2008. [6] M. J. Burge and K. W. Bowyer, Handbook of Iris Recognition. New York, NY, USA: Springer-Verlag, 2013. [7] Information TechnologyBiometric Performance Testing and Reporting Part 1: Principles and Framework, ISO/IEC Standard 19795-1:2006, 2006. [8] A. Lanitis, A survey of the effects of aging on biometric identity verication, Int. J. Biometr., vol. 2, no. 1, pp. 3452, 2010. [9] N. Ramanathan, R. Chellappa, and S. Biswas, Computational methods for modeling facial aging: A survey, J. Vis. Lang. Comput., vol. 20, no. 3, pp. 131144, Jun. 2009. [10] Germanys Federal Ofce for Information Security, Evaluation of ngerprint recognition technologiesBionger, Bundesamt fur Sicherheit in der Informationstechnik, Fraunhofer IGD, Darmstadt, Germany, Public Final Rep. 1.1, 2004. [11] L. Flom and A. Sar, Iris recognition system, U.S. Patent 4 641 349, Feb. 3, 1987. [12] J. Thornton, M. Savvides, and V. Kumar, A Bayesian approach to deformed pattern matching of iris images, IEEE Trans. Pattern Anal. Mach. Intell., vol. 29, no. 4, pp. 596606, Apr. 2007. [13] K. Miyazawa, K. Ito, T. Aoki, K. Kobayashi, and H. Nakajima, An effective approach for iris recognition using phase-based image matching, IEEE Trans. Pattern Anal. Mach. Intell., vol. 30, no. 10, pp. 17411756, Oct. 2008. [14] D. Monro, S. Rakshit, and D. Zhang, DCT-based iris recognition, IEEE Trans. Pattern Anal. Mach. Intell., vol. 29, no. 4, pp. 586595, Apr. 2007. [15] Iris Recognition, Wikipedia. (2011, Mar. 17) [Online]. Available: http://en.wikipedia.org/wiki/Iris_recognition [16] P. Tome-Gonzalez, F. Alonso-Fernandez, and J. Ortega-Garcia, On the effects of time variability in iris recognition, in Proc. IEEE Int. Conf. Biometr. Theory, Appl. Syst., Oct. 2008, pp. 16. [17] L. Masek and P. Kovesi, MATLAB source code for a biometric identication system based on iris patterns, B.E. thesis, School Comput. Sci., Softw. Eng., Univ. Western Australia, Perth, WA, Australia, May 2003. [18] S. Baker, K. W. Bowyer, and P. J. Flynn, Empirical evidence for correct iris match score degradation with increased time-lapse between gallery and probe matches, in Proc. Int. Conf. Biometr., 2009, pp. 11701179. [19] P. J. Phillips, W. T. Scruggs, A. J. OToole, P. J. Flynn, K. W. Bowyer, C. L. Schott, and M. Sharpe, FRVT 2006 and ICE 2006 large-scale experimental results, IEEE Trans. Pattern Anal. Mach. Intell., vol. 32, no. 5, pp. 831846, May 2010. [20] D. Rankin, B. Scotney, P. Morrow, and B. Pierscionek, Iris recognition failure over time: The effects of texture, Pattern Recognit., vol. 45, no. 1, pp. 145150, 2012. [21] N. Sazonova, F. Hua, X. Liu, J. Remus, A. Ross, L. Hornak, and S. Schuckers, A study on quality-adjusted impact of time lapse on iris recognition, Proc. SPIE, vol. 8371, pp. 83711W-183711W-9, Apr. 2012. [22] Neurotechnology. (2013, Feb. 4). VeriEye SDK, San Francisco, CA, USA [Online]. Available: http://www.neurotechnology.com/verieye.html [23] S. P. Fenker and K. W. Bowyer, Experimental evidence of a template aging effect in iris biometrics, in Proc. IEEE Comput. Soc. Workshop Appl. Comput. Vis., Jan. 2011, pp. 232239. [24] S. P. Fenker and K. W. Bowyer, Analysis of template aging in iris biometrics, in Proc. IEEE Comput. Soc. Workshop Biometr., Jun. 2012, pp. 4551. [25] M. E. Schuckers, Computational Methods In Biometric Authentication. New York, NY, USA: Springer-Verlag, 2010. [26] A. Czajka, Template Ageing in Iris Recognition. Barcelona, Spain: BioSignals, 2013. [27] C. A. P. Cavallotti and L. Cerulli, Age-Related Changes of the Human Eye. Totowa, NJ, USA: Humana Press, 2008.
273

A change in the shape of the cornea can cause a change in the image of the iris texture. A change in eyelid position can cause a change in the amount of iris texture imaged. A change in dilation can cause a change in the iris texture that appears in the image. In all of these cases, the texture of the iris surface itself may remain the same, and a template aging effect can still occur.
4) PERCEPTION OF STABLE IRIS TEXTURE OVER TIME DOES NOT IMPLY CLOSENESS OF BIOMETRIC MATCH

In one experiment, Hollingsworth et al [40] showed subjects pairs of iris images that were either from identical twins or from unrelated persons. In another experiment, the pairs of images were either from the left and right eyes from the same person, or from unrelated persons. In both cases, subjects were accurate at categorizing pairs of images that belong together versus pairs that do not. This shows that humans can perceive a similarity in the iris texture between left and right eyes of the same person, or between eyes of identical twins. However, using the same images, the average iris recognition score was no closer for identical twins, or for the left and right eyes of same person, than it was for unrelated persons. This shows that humans perceive similarities in iris texture that are not reected in iris recognition match scores. Thus, human perception of a stable iris texture pattern over time does not necessarily imply anything about iris recognition template aging. Studies of iris biometric template aging should be done in the context of recognition match scores, not human perception of texture similarity.
5) IRIS TEMPLATE AGING IS OBSERVED OR MEASURED EXPERIMENTALLY

An estimate of the size of iris template aging is necessarily made in the context of a particular combination of sensor, subject dataset, time interval and matcher version. Varying any one of these elements will generally affect the estimated template aging effect. Additionally, an image dataset of poorer quality may make it harder to experimentally observe a template aging effect [21]. Similarly, a matcher of poorer quality may make it harder to experimentally observe a template aging effect [26].
Acknowledgment

The term rolling re-enrollment was suggested by John Daugman.

1 The term rolling re-enrollment is due to John Daugman.
VOLUME 1, 2013

S. P. FENKER et al.: Template Aging Phenomenon

[28] PubMed Health, U.S. National Library of Medicine. (2013). Eyelid Drooping [Online]. Available: http://www.ncbi.nlm.nih.gov /pubmedhealth/PMH0002013/ [29] J. R. Price, T. F. Gee, V. Paquit, and K. W. Tobin, On the efcacy of correcting for refractive effects in iris recognition, in Proc. IEEE Conf. Comput. Vis. Pattern Recognit., Jun. 2007, pp. 16. [30] L. R. Kennell, R. P. Broussard, R. W. Ives, and J. R. Matey, Preprocessing of off-axis iris images for recognition, Proc. SPIE, vol. 7119, pp. 7119061711906-9, Sep. 2008. [31] H. J. Santos-Villalobos, D. Barstow, M. Karakaya, E. Chaum, and C. Boehnen, ORNL biometric eye model for iris recogition, in Proc. IEEE 5th Int. Conf. Biometr., Theory, Appl. Syst., Sep. 2012, pp. 176182. [32] K. Hayashi, H. Hayashi, and F. Hayashi, Topographic analysis of the changes in corneal shape due to aging, Cornea, vol. 14, no. 5, pp. 527532, 1995. [33] F. S. Vihlen and G. Wilson, The relation between eyelid tension, corneal toricity, and age, Investigat. Ophthalmol. Vis. Sci., vol. 24, no. 10, pp. 13671373, 1983. [34] D. A. Atchison, E. L. Markwell, S. Kasthurirangan, J. M. Pope, G. Smith, and P. G. Swann, Age-related changes in optical and biometric characteristics of emmetropic eyes, J. Vis., vol. 8, no. 4, pp. 120, Apr. 2008. [35] J. Birren, R. Casperson, and J. Botwinick, Age changes in pupil size, J. Gerontol., vol. 5, no. 3, pp. 216221, 1950. [36] B. Winn, D. Whitaker, D. Elliot, and N. Phillips, Factors affecting lightadapted pupil size in normal human subjects, Investigat. Ophthalmol. Vis. Sci., vol. 35, no. 3, pp. 11321137, 1994. [37] P. Bitsios, R. Prettyman, and E. Szabadi, Changes in autonomic function with age: A study of pupillary kinetics in healthy young and old people, Age Age., vol. 25, no. 6, pp. 432438, 1996. [38] K. Hollingsworth, K. W. Bowyer, and P. J. Flynn, Pupil dilation degrades iris biometric performance, Comput. Vis. Image Understand., vol. 113, no. 1, pp. 150157, Jan. 2009. [39] P. Grother, E. Tabassi, G. W. Quinn, and W. Salamon, NIST IREX I: Performance of iris recognition algorithms on standard images, Information Access Division, NIST, Gaithersburg, MD, USA, Interagency Rep. 7629, Sep. 2009. [40] K. Hollingsworth, Genetically identical irises have texture similarity that is not detected by iris biometrics, Comput. Vis. Image Understand., vol. 114, no. 11, pp. 14931502, 2011. [41] P. Grother, G. W. Quinn, J.R. Matey, M. Ngan, W. Salamon, G. Fiumara, and C. Watson, IREX III: Performance of Iris Identication Algorithms, Information Access Division, NIST, Gaithersburg, MD, USA, Interagency Rep. 7836, Apr. 2012. [42] AD 100 Iris Camera, Iris Guard (2013, May 15) [Online]. Available: http://www.irisguard.com/ [43] E. Ortiz and K. W. Bowyer, Dilation-award multi-image enrollment for iris biometrics, in Proc. Int. Joint Conf. Biometr., Washington, DC, USA, 2011, pp. 17. [44] E. Ellavarason and C. Rathgeb, Template ageing in iris biometrics: An investigation of the ND-iris-template-ageing-2008-2010 database, Biometrics and Internet-Security Research Group, Center for Advanced Security Research, Darmstadt, Germany, Tech. Rep. HDA-da/sec-2013-001, 2013. [45] A. Uhl and P. Wild, Weighted adaptive hough and ellipsopolar transforms for real-time iris segmentation, in Proc. 5th Int. Conf. Biometr., 2012, pp. 283290. [46] S. Baker, K. W. Bowyer, P. J. Flynn, and P. J. Phillips, Template aging in iris biometrics: Evidence of increased false reject rate in ICE 2006, in Handbook of Iris Recognition, M. Burge and K. W. Bowyer, Eds. New York, NY, USA: Springer-Verlag, 2013.

SAMUEL P. FENKER received the B.S. degree (cum laude) in computer science from the University of Notre Dame, Notre Dame, IN, USA, in 2012. As an Ateyeh Undergraduate Research Scholar during his career at Notre Dame, he has co-authored and presented several works focusing on the effects of template aging in iris biometric systems. He currently works in research and development for a major electronic health record vendor.

ESTEFAN ORTIZ is currently pursuing the Ph.D. degree in iris biometric systems from the Department of Computer Science and Engineering, University of Notre Dame, Notre Dame, IN, USA. He received the Master of Science degree in electrical engineering from the University of Hawaii at Manoa, Manoa, HI, USA, in 2006, and the Bachelor of Science degree in electrical engineering and the Bachelor of Arts degree in mathematics from St. Marys University, San Antonio, TX, USA, in 2003. He is a Deans Fellow with the University of Notre Dame. He has served as a Principal Research Engineer with the Research Corporation, University of Hawaii. His current research interests include biometric systems, machine learning, computer vision, and pattern discovery.

KEVIN W. BOWYER is currently the SchubmehlPrein Professor and a Chair of the Department of Computer Science and Engineering, University of Notre Dame, Notre Dame, IN, USA. He has served as an Editor-in-Chief of the IEEE TRANSACTIONS ON PATTERN ANALYSIS and MACHINE INTELLIGENCE and the IEEE Biometrics Compendium. He is the founding General Chair of the IEEE International Conference on Biometrics Theory Applications and Systems series, having served as a General Chair of the BTAS in 2007, 2008, and 2009 conferences. He served as a General Chair of the 2011 International Joint Conference on Biometrics, and as a Program Chair of the 2011 Automatic Face and Gesture Recognition conference. He is a fellow of the IAPR and a Golden Core Member of the IEEE Computer Society. His latest book is the Handbook of Iris Recognition. He received the Ph.D. degree in computer science from Duke University, Durham, NC, USA.

274

VOLUME 1, 2013