NO2ID

[Just posted to the Albinism Fellowship]

You are probably aware that the government proposes introducing a system of universal ID cards and population registration. I'm writing to ask if anyone here knows about (or has been approached by the Home Office about) the practicability and/or problems of Iris-pattern recognition for albino eyes.

Iris scans are one of the preferred biometrics that the Home Office proposes to use in its national identity register to tie each individual to their government record. (There's an account by its inventor here: http://www.cl.cam.ac.uk/~jgd1000/iris_recognition.html)

While iris scanning is probably the best system of biometrics available it is by no means perfect (pace Prof. Daugman). After some harassment, the government has already acknowledged that it has problems with picking up patterns in heavily pigmented eyes, and claims to be recalibrating the systems to cope better with black people. But how about the reverse? How well does the system work on the smaller minority lacking pigment? The illumination used rarely causes discomfort and is not known to be injurious to normal eyes, but is this still the case for albino eyes.

I don't know the answer. I hope you do. But I bet the Home Office hasn't yet thought of it.

I haven't been approached by the Home Office but I thought I'd have a look at the issue, as it's somewhat interesting.

First, albinism affects around 1 in 17,000 persons. This is greater than the proportion of persons with aniridia (lack of iris, which renders iris recognition ineffective), which affects around 1 person in every 56,000. Thus, any problems affect only a very small minority of persons. That, however, does not mean health and safety problems and biometric performance problems should be ignored - though there may well be more pressing problems that rightly deserve greater and more immediate attention.

Concerning aniridia, the solution is to use other biometric modalities, such as fingerprint, hand geometry, hand vein patterns, etc.

In fact, multi-modal and multi-instance (ie 2+ fingerprints) biometrics are required to provide adequate performance for Detection of Multiple Applications (DMA). So the availability of multi-biometrics is not just a requirement for those without irises.



There are techniques in statistical pattern matching, well-known to experts in the field, concerning recognition score normalisation for each enrolled person. There is also even better stuff, that people like me have been working on, that is perhaps not so well known.

There are, with albinos, also sometimes the problems of Nystagmus (irregular rapid movement of the eyes back and forth) and Strabismus (muscle imbalance of the eyes leading to "crossed eyes" or "lazy eye" (thanks Wikipeadia). Both of these would adversely affect capture of iris images, except with more expensive equipment that uses video capture rather than still image capture of posed images.



Iris recognition does not (so I understand from John Daugman) work for those with aniridia.

For those with less pigment in the infra-red range of wavelengths, there will very likely be lower performance. However, I understand that that blue eyes have more variability and contrast than brown, even though they have less melanin pigment.

I don't know about albino eyes. However, it is almost certainly better to exploit what discrimination there is between their irises than to ignore it, given that the equipment would be in place for the other 99.9941% of the population. This is certainly the case for DMA. The case for using iris at Points of Use (PoUs) for personal identity verification (with its non-zero capture time) of those with albinism would need to be made specifically for each PoU, or PoU type, on the basis of a full security assessment.



On this, it is worth noting that iris recognition uses low power near infra-red illumination from Light-Emitting Diodes (LEDs).

John Daugman writes:



Thus, the NIR illumination is not visible to humans; it therefore seems most doubtful that it is a problem for albinos. That is unless they have additional sensors to those in non-albinos, rather than more sensitive sensors to the same visible wavelengths.

However, this is not certain. We can go a little further into this without great expense. Acquire an albino volunteer and subject him to NIR illumination from a typical TV remote control unit. I've just done this to myself and could not see or feel anything, even at a range of about 2 inches; this was from 2 different IR controls, both known to be working immediately before and afterwards. However, it would be prudent to start at a fair distance, just in case other people do have greater sensitivity than me.

Typical TV/audio IR controls work at around a wavelength of 940nm (http://en.wikipedia.org/wiki/Image:Control-remote-spectrum.png), which is greater than the 850nm and 900nm, which is, I understand, most typical of iris recognition.

However, I think this sort of experiment might go some way towards identifying whether there is a health/safety problem for albinos, even if it does give firm confirmation that there is no problem, through the wavelength difference of around 10%.

Best regards

Apologies: correction here to the last 2 paragraphs in my posting immediately above.





Best regards

Thank you Nigel -

Very helpful.

I suspect the latter is the case, in that lack of pigment reduces protection. Many albinos have trouble with normal daylight (which is of course considerably more powerful than the LED illumination) in most circumstances. I have lost contact with my one albino friend some years back, so I can't introduce you to a volunteer. (Though you might need more than one, since people always vary.)

* Note for the less-technical even than me: The "NIR" referred to in this case is the Near Infra-Red No doubt parties should have a policy on it, too.

Nigel,

You quoted John Daugman as saying



Could you expand on this. Why does the imaging need to be invisible to humans? Also, are there other technical reasons relating to the imaging - maybe to gain a picture with better contrast - or is this the sole reason?

By the way, your experiment with the remote control isn't necessarily all that useful; the LEDs are pulsed (in order to preserve battery life), so although the instantaneous power is high enough to outshine natural sources of near infra-red in the room, the average is possibly quite low (it depends on the carrier frequency, the mark/space within the carrier, the coding scheme, and the repetition rate when the key is held down).

Jonathan Clift wrote several things:



Firstly, this thread started off about albinism, on which there did not seem to be any easily available information on the implications for iris recognition. Therefore, I thought it would be useful to put together what I could find in one place.

Extending the thread to the whole issue of iris recognition strikes me as unnecessary. There is a lot of information, including the very readable 10-page article by John Daugman How Iris Recognition Works that I linked to in my first comment on this thread. Secondly, there is a vast about of other information on iris recognition on the rest of John Daugman's personal webpages at Cambridge University; they too are very readable, considering the overall complexity of the technology: it's a skill he has. Thirdly, there is a ton of stuff available through use of WWW search engines.

My personal skills in iris recognition are much more limited, and are currently based largely around the mathematics of statistical pattern matching. See here, but most NO2ID people will not find this particularly easy to digest (the phrase "near-impossible" springs to mind). Nevertheless, it is useful: it helps improve technical performance, thus making iris recognition and other biometrics more effective in providing additional security through personal identification (in so far as that itself is actually useful).

Diatribe nearly over. Now back to Jonathon's questions.

I don't actually know whether my attempts at answers are actually correct, and I'm certainly not going to search our reliable references that cover the points. So please use your own judgement in whether what I write sounds reasonable.



On the technical side, visible illumination significantly above the ambient would cause the pupil to contract. This is actually useful for iris recognition, as it makes the iris bigger and so measurements more reliable. However, the time taken for the pupil to contract and the stability of pupil size in somewhat beamed illumination (rather than a steady broadly diffused background illumination) might cause image capture difficulties: these points would apply especially to lower-cost equipment that uses single shot still/posed photos rather than selection from continuous video capture.

On the non-technical side, people are more likely to dislike illumination they can see than illumination they cannot perceive at all. It's even possible that John Daugman took into account the light-sensitivity of albinos in his choice of illumination, though I cannot remember reading that anywhere.



I understand that the image variability of brown eyes in visible light is not very great, and that it is much greater in infra-red light. I recollect that eye colour is largely/somewhat Mendelian in character and that in Caucasians the "brown gene" is dominant; also several other ethnic types have only brown eyes. Thus there is a preference for coping with brown eyes, even if blue eyes do worse in infra-red than in visible light (and I don't know on that). There is more information on eye colour in Wikipeadia, but it does not cover iris recognition and infra-red illumination effects.



Well, I was trying to be helpful on the issue of albinism. I think the IR illumination used in iris recognition is also usually applied only during image capture. I don't know the details and suspect it varies with manufacturer. From personal recollection using iris recognition, I suspect the illumination is on for around one second for each capture: there is certainly usually a visible LED that indicates "pose now".

On TV infra-red controls, I note Jonathon's poo-pooing of my idea. I'm not convinced that he is right: effectively saying that TV remote controls would not be visible to humans through the average power and duty cycle if they operated at a visible light wavelength. Having bashed a few Wikipeadia pages, average power over the on cycle part seems typically in the range 100mW to 1W; main on-periods are typically of 10s of milliseconds and overall duty cycle is greater than 25% and usually higher. My controls repeat transmit around 5 times per second (from the flashing of the receivers' LEDs). Also, as we know from flash photography, peak illumination is at least as much of a problem as average illumination. However, the remotes do vary a lot between manufacturers, in these aspects and also in modulation method (which matters too).

If any readers (or their friends) have albinism with some light sensitivity, and would like to contribute to a simple, if not conclusive, experiment on albino sensitivity to infra-red illumination as used by iris recognition devices, please get in touch with me (ncs AT camalg.co.uk) fairly soon, and I see what I can arrange: a good bet would be the Biometrics 2006 exhibition in London on 19th/20th October.

Best regards

Thanks for your reply, Nigel, and sorry for going off at a bit of a tangent. The material on John Daugman's site is very interesting (thanks for bringing it to my attention) but he concentrates on the theory and algorithms and, understandably, doesn't have much to say on the actual implementation. The only thing that seemed relevant was a picture of a brown iris under infra-red where the caption suggested that some of the features were enhanced. I thought that with you knowing a fair bit about all this you might have been able to give a quick answer (I certainly wasn't expecting you to do all my reading and research for me). In future I shall Google first and ask later.

I wasn't trying to expand the thread to "the whole issue of iris recognition" but the butterflies of my mind were having one of their more chaotic days and I had started thinking about eye hazards; whether albinism might expose someone to a problem where none existed otherwise; and I was slightly concerned at your encouraging people to experiment in a situation where the aversion response they would naturally have with visible light was circumvented. Having thought about it more, I don't think it's really an issue since albanism seems to result in a translucent rather than transparent iris and that obviously diffuses any light passing through and obviates any additional problem that people with this condition might have (at least in terms of safety with LED illumination), though your experiment to determine any discomfort is still valid. It still might be as well, if you do this experiment, to stick to the normal use of the scanner and not encourage people to place their eyes very close to the source of illumination.

As to your TV remote-control, I wasn't "poo-pooing" your idea but trying to be helpful by pointing out that it might not be a very good analogue of the scanner illumination. If you think it is, then fine. I didn't say anything about what I thought the equivalent in visible light might look like - if you want me to guess I would say visible but not uncomfortably bright; certainly not the "not visible to humans" that you have me "effectively" saying.

BTW your own work looks interesting, but, as you surmised, it's way over the top of my head and I certainly won't be discussing the maths with you in the foreseeable future.

Another reason for using NIR 700nm..900nm wavelengths is that melanin has low absorption at these wavelengths, since melanin is the main contributer to dark pigmentation it makes the iris recognition technology quite robust to differences in pigmentation. How Iris Recognition Works