|
|
As with almost every new technology, iris recognition has both the potential
to be a convenience enhancer or an obstacle. Dr John Daugman, researcher
in statistical pattern recognition, computer vision, decision theory,
and computational neuroscience at Cambridge University, explains.
Humans have traditionally identified each other by their appearance,
by the sound and content of their speech, and by context. If the other
person is neither visible nor audible for instance when receiving
their email we either simply accept their asserted identity, allow
it to establish itself by shared knowledge and context, or rely on special
secret knowledge such as encryption keys.
Identification among strangers in official interactions, such as immigration
passport control or financial transactions, has traditionally relied upon
special possessions (documents such as passports and identity cards),
or secrets (eg passwords).
With the arrival of automation, identification of persons has continued
to rely largely on special possessions (magnetic or optical stripe cards)
and on secrets (computer login password, or cashpoint PIN number).
These methods go back thousands of years: Bronze Age priests had special
amulets to establish their identity, and the Roman centurions used secret
military passwords.
Traditional identification
However, the problem with identification based either on special possessions
or on secret knowledge is that it only really establishes that the possession
or the secret was present at the transaction. It does nothing to confirm
that the holder of the special possession or of the secret was the rightful
person.
This fundamental problem with traditional methods of identification
whether between human strangers, or identification of persons by machines
took on new urgency on September 11, 2001.
Some of the terrorist hijackers who crashed the aeroplanes into buildings
were on FBI 'watch lists', but they were using false passports. With no
reliable mechanism to confirm any connection between a passport (or other
identity device) and its holder, September 11 was a catastrophe waiting
to happen.
Behavioural biometrics
The technologies called biometrics (from biological measurement) seek to
tie identity much more tightly to a person's particular unique features.
These could be anatomical, physiological, or even behavioural.
The sound of a person's voice, or the way in which they sign their name,
are examples of behavioural biometrics. Their blood type, or markers in
their tissue or fluid samples (including DNA itself) are examples of physiological
biometrics, although these would be used more typically in forensic applications
rather than in real-time, live applications.
Most currently used biometrics are anatomical: facial appearance, hand geometry,
fingerprints, retinal vein patterns, and iris patterns.
This article aims to explore biometric issues of particular relevance to
persons with disabilities. Will biometric identification mean more obstacles
and exclusion? Or could biometrics actually be facilitators, and convenience
enhancers?
As the inventor of iris recognition, I am particularly concerned to ensure
that the latter and not the former is the upshot of the use of the iris
biometric. But these issues apply equally to all biometric identification
systems.
As developers of such systems we need to solicit feedback and advice from
persons with disabilities to ensure the maximum benefit for all from such
technologies, and particularly to ensure that these technologies reduce,
rather than add to exclusion.
I hope that the issues raised in this article will contribute to that goals.
Fundamentals of iris recognition
First let me say a bit about iris recognition and how it works. The iris
is an internal organ of the eye perhaps the only internal organ of
the body that is routinely visible from outside and its patterns are
resolvable with good videocameras from distances of up to about a metre.
The iris is located behind the cornea of the eye, and behind the aqueous
humour, but in front of the lens. Its only physiological purpose is to control
the amount of light that enters the eye through the pupil, by the action
of its dilator and sphincter muscles that control pupil size, but its construction
from elastic connective tissue gives it a complex, fibrillous pattern.
Its morphogenesis begins in the third month of gestation and is largely
complete by the eighth, although changes in pigmentation can occur during
the first year of life (hence most babies are born with slate-blue eyes,
regardless of their ultimate, genetically determined eye colour).
There is no genetic determination of the detailed iris texture, as sighted
readers can confirm just by examining the detailed texture in their left
and right eyes (which are, of course, genetically identical).
Apart from the occasional appearance of freckles or other pigmentation changes
caused by some eyedrop treatments for glaucoma, there is no evidence for
any change of iris pattern over a person's life.
Iris patterns have a high degree of randomness in their structure. This
is what makes them unique. Every biometric depends upon random variation
among different persons in the chosen measurements, in order to guarantee
that a particular pattern is unique to just one person and thus can serve
as a reliable automatic identifier of them.
The greater the degree of randomness, the greater the likelihood of uniqueness.
An iris pattern is a bit like a plate of 250 sticky noodles, thrown down
to create a random pattern, and so the combinatorics of possible patterns
that the pectinate ligament in the trabecular meshwork of the iris can form
is truly astronmical.
My algorithms encode the iris pattern into an abstract mathematical description
called an 'IrisCode'. This process relies upon two-dimensional wavelets
(mathematical functions that are like restricted Fourier components, ie
sinewaves multiplied by Gaussian envelopes to give them locality). The result
of the wavelet analysis is that any piece of an iris can be said to have
a certain phase. This 'phase sequence' allows an iris pattern to be encoded
in a total of 512 bytes worth of information.
Making a match
Whenever a person presents his/her eye to a camera, its IrisCode is computed
within a second or less, and then this is compared with all previously enrolled
IrisCodes in the relevant database to see whether any of them match.
An important point is that the person does not need to assert any identity;
the algorithms are powerful enough (and fast enough) to discover their
identity, if they have been seen before and enrolled. The speed of the
database search is about 100,000 IrisCodes per second.
This ability to be recognised without having first to assert an identity
eg by swiping a card, or by typing in a name or a PIN number
is one potential advantage of iris identification for persons who have
limited use of arms or hands.
This 'hands-free' use of iris recognition is possible because the probability
of false matches is so low that the algorithms can 'afford' to search
large databases exhaustively, rather than just answering a single yes/no
question about a claimed identity.
In many millions of IrisCode comparisons that have been done in tests
by independent laboratories (eg the UK's NPL Labs), so far there has never
been a single false match reported.
The potential limitations of iris recognition for persons with various
disabilities include the following:
- a person must of course have an eye, with an iris. According to the
US National Eye Institute (www.nei.nih.gov),
the condition of aniridia (lack of an iris) occurs in 1.8 of 100,000
births. Because it is genetically linked, the condition usually affects
both eyes according to the RNIB (www.rnib.org.uk/info/aniridia.htm),
but its incidence covers a wide spectrum of partial conditions such
as just chronically enlarged pupils. Iris recognition requires the pupil
to have a diameter less than about 75% of the iris.
- blind persons may have difficulty in getting themselves aligned with
the iris camera at arm's length, because some such systems rely on visual
feedback via a mirror or LCD display to guide the user into alignment
with the camera. (Other more sophisticated iris cameras are mounted
on automatic pan and tilt platforms that actively home in on an eye,
including autozoom and autofocus, so very little cooperation from a
user, or indeed vision, is needed.)
- persons with pronounced nystagmus (tremor of the eyes) may have difficulty
in presenting a stable image; however, some iris cameras now use stroboscopic
(flashed infrared) illumination with very fast camera integration times,
on the order of milliseconds, so tremor becomes unimportant for image
capture.
Within reach?
Perhaps the most important disability issue involving iris recognition arises
with wheelchair-bound people, because a wall-mounted iris camera presumes
that a person's head is within a particular range of heights. All fixed
cameras are swivel mounted to adjust for height, but still their 'capture
box' is limited to about 18ins in height variation.
Wheelchair access requires either that the entire unit can move up or down
(eg on a sliding pole, as used in the EyeTicket installations at airports
such as Heathrow and Charlotte www.eye
ticket.com), or else that a tethered, handheld camera be used which
a person picks up like a telephone handset and brings to the appropriate
level of their eyes, within arm's length. Such handheld iris cameras are
made by Panasonic (an example can be seen at www.panasonic.com/medical_industrial/iris.asp).
As with almost every new technology that seeks to find its place in everyday
life, iris recognition has both the potential to be a convenience enhancer
(including an access enhancer), but also the potential to be an obstacle
or excluder if improperly configured or installed without consultation and
guidance from disabled persons.
Because it allows hands-free, automatic, rapid and reliable identification
of persons, it can facilitate access for persons unable to engage in the
standard mechanical transactions of access. But it must not presume universal
uniformity among persons and their bodies. The variability of people is,
after all, the heart and soul of biometric technology.
For further information: www.CL.cam.ac.uk/users/jgd1000/
|
