CLPEX Chat Board

In a recent article (see Detail # 313 UK news article) the following text appears:

Last month, Cambridge University IT expert Professor John Daugman warned that using fingerprints to identify individuals meant one in a thousand card-holders would be wrongly matched with other people's details.

He said: "I don't want to be very pessimistic and say this whole system is doomed to fail but I will say that just from analysing the mathematical requirements, if it is just fingerprint, it is unlikely to be able to succeed."

The Conservatives say that, by the time the entire UK population is on the ID cards database, there could be 40,000 "false matches".

Shadow home secretary David Davis said: "Gordon Brown pretends ID cards will secure our identities, but tens of thousands of people could be falsely accused of not being who they are.

"That will create, not solve, problems and divert time, money and effort from vital areas of public protection, including at our borders and in the fight against terrorism."

But a Home Office spokesman said it was "not true" to suggest the database could not cope with vast records.

In the US, the FBI had a system with 47 million records and the US-Visit scheme - which records fingerprints of those arriving in the country - held details from more than 80 million people, he added.

It may be obvious to most folks, but I wanted to point out something we should all be aware of. The "False matches" Daugman is talking about are non-matches returned by the system where there exist matches at that same score. Remember, Daugman is an iris man, and there is hardly any overlap between match and non-match scores. What he isn't accounting for is the human examination that occurs to resolve those questionable scores to matches and exclusions. So when we see that there could be 40,000 "false matches" per year in the UK, that would be a workload of at least 40,000 RESOLVES for human examiners to conduct. The next statement completely disregards this fact - 40,000 people would NOT be accused of being someone they are not. They wouldn't even know the system flagged their identity and an examiner said "exclude".

It is difficult to know whether this statement by Daugman was taken out of context or whether this is a snipe at fingerprints, since he is an iris guy. It is very difficult to imagine that he wouldn't know that fingerprint systems have humans in the loop. But just in case an attorney brings up a recent article involving 40,000 false matches a year, now you have your source... and your ammo.

Keep up the good fight,

-Kasey

I talked to John Daugman this morning, and asked him about Kasey’s posting. He said that the statistic he was quoting (1,000 False Matches in a two-index-finger based system once the enrolled database reaches 6,000,000) comes from: "Matching Performance for the US-VISIT IDENT System Using Flat Fingerprints" -- May 2004. There it states the statistic FMR=0.1% with TAR=95% on a database of 6,000,000 for the case of matching two fingers at once (L and R index fingers). He also points out that a more detailed review of fingerprint accuracy and the distributions obtained for different NFIQ quality scores was published recently in the Proceedings of the IEEE (November 2006) by 3 people from NIST.

I am not planning to get into this discussion, but thought it may be helpful to convey this information.


Itiel