Interview: John Daugman
22 June 2016 13:55 GMT

Interview with John Daugman, Professor of Computer Vision and Pattern Recognition at Cambridge University, UK.   He is the inventor of iris recognition, and his algorithms are the basis of all public deployments of this biometric technology around the world.

Did you expect the growth we see today when you originally developed and patented the first algorithms for iris recognition?

The main surprise was in the demographics. I had expected uptake in Northern European countries whose indigenous populations tend to have blue eyes with very salient and striking iris patterns.   But instead of Scandinavia, it was Japan, Korea, Malaysia, some Gulf States and other areas with predominantly "dark brown eyes" that acquired the first licenses.   The reason as explained to me by a Japanese licensee was: "We are amazed that the iris can be used in this way, because in the visible wavelengths our own iris just looks dark and uniform, we cannot see any patterns, so they are like a surprising hidden secret!" 

The use of invisible (near-infrared) light in iris cameras extracts very rich, almost lunar and cratered, texture from dark brown eyes whose owners were not aware they had it.

Hence the unexpected demographic enthusiasm for the technology.

Today more than a billion persons have used it, about 95% having dark brown eyes.

Can you explain why entropy and parallel bit logic played such an important role in the iris aspect of India's Aadhaar project?

Entropy is the fundamental concept in Information Theory, which to my surprise is rarely discussed in biometrics.   Entropy is what determines collision avoidance – resistance to False Matches– in the same way that the keylength (in bits) of a cryptographic key determines its strength.

Iris patterns, and their encoded IrisCodes, contain much more entropy than do weaker biometrics.  That is why iris recognition has such legendary resistance to False Matches. 

For example, NIST has confirmed after 1.2 trillion (1.2 million-million) iris comparisons that even if a match is declared between two IrisCodes which disagree in 28% of their bits, the measured False Match probability remains only 1 in 40 billion.

Now, THAT is what having large biometric entropy buys you.

Parallel bit logic exploits this entropy to deliver extremely fast matching because a very simple logical operation on large chunks of bits in IrisCodes, done in parallel all at the same time, measures their degree of similarity and makes a match decision.   The blisteringly high speed of iris matching in large database searches is the result of the bit-level parallelism of that simple logical operation.

Do you believe automatic iris recognition will be a good fit for high-volume, high-speed applications such as tracking multiple people passing through an airport?

If the application cares at all about avoiding False Matches, then obviously yes.   I am shocked about the uptake of face recognition in many airport deployments.   Face recognition is most optimistically benchmarked at a False Match Rate of 1 in a thousand.   That is appallingly poor, because as soon as 38 people are assembled together at random, it becomes more likely than not that at least one pair of faces would be confused with each other at that error rate. 

Next time you stand in a queue at Passport Control, turn around to look at the first 38 people standing behind you, and spot the pair who would be confused with each other at that error rate.  The mathematics is straightforward.  It is the same sort of calculation as the famous "birthday problem", which shows that as soon as 23 or more randomly chosen people are gathered in a group, it is more likely than not that at least one pair have the same birthday.

Do you think that, with implementations such as iris recognition at border control, the industry has progressed well in terms of public acceptance?

Public acceptance of iris recognition has been impeded by so much ignorance, especially in the press, about the eye and the iris (which is constantly confused with the retina, etc).

Statements are made that, for example, iris recognition will not work if you are pregnant.   These bizarre folk beliefs then become promoted into "facts", simply by repetition.

And it is not only the press. There is a professor in London who told the BBC that iris recognition would not work if you have astigmatism.

She just made it up, from assumptions, and from confusing the different parts of the eye like the cornea, the lens, the iris, and the retina. Iris recognition is affected by astigmatism about as much as it is by pregnancy.

But her highly publicised BBC interview turned it into a "fact".

What is your take on the iris-aging debate?

NIST conducted a major study of this involving 3.5 million iris images collected over 6 years from 622,464 subjects.   NIST also analysed operational border-crossing data from international airports, involving more than a million iris transactions over a nine-year period.  NIST concluded that there was no systematic effect.

On the other side, the academic labs promoting the aging claim have studied only a few hundred subjects, under varying acquisition conditions.  Nobody has ever presented any photographic evidence of changes in iris patterns over time, apart from injury or big changes in pupil dilation that are inadequately reversed in the algorithm.

It sometimes seems that the main promoters of the iris aging claim try to win the argument by large numbers of press releases and media interviews.  And in their latest academic paper, there is an obvious error by a factor of 14 in one of their key calculations, leading to a 14-fold over-estimation of an effect.   But the popular media just splash the story as enthusiastically as the one about the negative impact of pregnancy on iris recognition.

Related articles

Citigroup and Diebold reveal iris-scanning ATM plans
Aadhaar architect foresees iris scanners on smartphones
BAE demonstrates face and iris recognition

Industry Events

connect:ID 17-18 Mar 21