But do the eyes really have it?
The Statesman
11 Jul 2013
Usha Ramanathan
In September 2012, two years after enrolment had begun, the UIDAI produced a report on iris authentication. As in the proof of concept (PoC) on fingerprint authentication, the iris report too was about field-testing the technology, and not a scientific study. This allowed for cleansing the data "of exceptions and anomalies", checking out vendors and their devices, encountering the people who came in their infinite variety - those with squints, those who had undergone eye surgery, those who had eye deformities and those without sight. The PoC was done in a semi-urban taluka in Mysore over two months in 2012 with 5747 residents. As with the fingerprint report, here too the percentages that the UIDAI records are intended to reassure, but the devil is in the detail.
The older population, those who have undergone surgeries, those unable to open their eyes wide, those with eye deformities and, especially those who had undergone cataract surgery using older techniques were expected to have trouble authenticating. But, it was said, while iris authentication is significantly improved by using the dual eye camera, those with a squint would be better off with a single eye camera. What effect there would be on the error rate as the database grows larger and larger is not reckoned with.
Yet, these concerns lose their urgency when viewed against the first presumption on which the PoC is based. "The iris does not get worn out with age, or with use," it says. "In addition, iris authentication is not impacted by changes in the weather." This seems an improbable claim, for it is difficult to imagine a part of the human body which withers with neither age nor clime. Still, the improbable is not necessarily the impossible.
This, the report claims, is a presumption drawn from iris technology literature. But, in a paper presented at the IEEE Computer Society Biometrics workshop on 17 June 2012, two professors from the Department of Computer Science and Engineering at the University of Notre Dame found something quite different. Samuel E Fenker and Kevin W Bowyer did a study of iris images acquired between 2008 and 2011 using state-of-the-art technology, with 322 subjects ranging from 20 to 64 years, 177 male and 145 female, of different races. In introducing their study, they explained that the prevailing view that iris is "essentially immutable over a person's life" had been repeated in several research papers, even though "we know that no studies with experimental results that support the conclusion that template ageing does not occur for iris biometrics" exist. This includes Daugman's 1994 iris biometrics patent which asserted that "the iris of every human eye has a unique texture of high complexity, which proves to be essentially immutable over a person's life." Fenker and Bowyer's paper was "the most extensive experimental investigation to date on template ageing for iris biometrics."
In brief, their study found "clear and conclusive evidence that template ageing does occur in iris biometric matching. Specifically, the experimental evidence indicates that the false non-match rate increases with increasing time between acquisition of the enrolment image and the image to be recognised." That is, as time elapses, the image alters from how it was when it was enrolled. "In our results," they said, "the false non-match rate increases by greater than 50 per cent with two years of time lapse." The 50 per cent indicates the rejection rate when it was sought to be authenticated, and it is disturbingly large.
Fenker and Bowyer are not biometric skeptics, and they offer ways to proceed once it is acknowledged that template ageing does occur for iris biometrics. One possible route is "that the user may simply need to be re-enrolled in the system after some determined period of time." Given that the drop in confidence in the biometrics occurs within just two years, re-enrolment is not even an option amidst the Indian population. And, they suggest, "once the fact that template ageing for iris biometrics is acknowledged, research effort may be focused on reducing the magnitude of the effect."
This is the state of knowledge in biometrics.
The iris authentication report recognises this when it says: "Few global initiatives have empirically published results on iris based online authentication in a context similar to aadhaar." It is this use of untested technology that has had critics of the project say that it is an experiment where India is the laboratory, and Indian residents are mere specimens.
Spoofing and fraud
It is not only the experimental stage of the technology that raises questions. It is also questions of spoofing and fraud.
On 30 September 2011 a meeting was held at the Planning Commission to discuss the issue of privacy. The UID project, and the Human DNA Profiling Bill which has in circulation since 2007 and which resurfaced more recently, prompted the meeting. Representatives from the UIDAI, Natgrid, the Department of Personnel and Training were present among others that included professionals and activists. J T D'Souza, a biometrics expert who is in the trade, was present, and he demonstrated fingerprint authentication done with a faked fingerprint made out of Fevicol and wax. It was his wife's fingerprint. It authenticated perfectly when he blew on the spoofed fingerprint to add moisture to its surface, so that the fingerprint reader could be made to believe that it was a live finger that was being applied to it. It is easy to spoof a fingerprint, he said. When it is cooperative, as it had been in his case where his wife gave her fingerprints willingly, he had used a plastic battery case into which he melted wax. When it had not quite set, the finger was pressed into the wax leaving an impression into which he poured Fevicol. When the Fevicol set, he had peeled it off and, hey presto, it was ready for use. When it is "non-cooperative", it may be an impression taken, say, from a glass or anything that is touched, the process would be a tad more tedious, involving using standard techniques from forensic sciences, making a positive, using a standard printed circuit board etching technique which is well known to any second-year electronic student or electronic hobbyist and use that as a template with Fevicol.
The danger is, too, that once the fingerprint has been compromised it cannot be changed, unlike a password or a pin number. In controlled spaces, biometrics may work because there are other controls along with the biometric. But a centralised database and long-distance authentication, D'Souza cautioned, is a prescription for fraud. D'Souza's demonstration of the use of the spoofed fingerprint to the students of a Bombay college is on youtube; there has been no reaction to it so far. At the Planning Committee meeting, the representatives of the UIDAI said they would look into it. Six months later when the report was released, there was no mention of this issue.
The problem is not only that it is an experiment, and just may fail. It is that what is being attempted is what Mr Nilakeni calls "doing government process re-engineering" with this experimental technology as its foundation.
(The author is an academic activist. She has researched the UID and its ramifications since 2009)
