JD is blogging about facial recognition software and attempting to identify terrorists. I commented that it's not ready for Prime Time yet, but thought I would expand on that just a bit.
As we know companies often make inflated claims about their products and this seems to be especially true in the field of software. Possibly because, for many years, too few people could prove them wrong. But Bruce Schneier a noted computer security expert with a most excellent monthly Cryptogram email does an excellent job of putting things in perspective.
From CFO.com we have the following:
Identix, a leading supplier of facial-recognition systems, claims that its equipment's accuracy rate can be as high as 99%. But Mr Schneier, the security expert, says that even with an accuracy rate of 99.99%, and assuming that one in 10m fliers is a suspect whose face is on the watch list, there will still be 1,000 false alarms for every suspect identified. And most terrorists are not on watch lists. Face-scanning may reassure people and may have a deterrent effect, but these meagre benefits do not justify the costs.
What isn't mentioned in the above quote, the lack of GOOD photographs of terrorists who are known. So, you skew the database even more because you are trying to match to a poor photograph. Terrorists, naturally, aren't going to go in front of a camera and have an excellent picture taken. And that's where we might get the best chance of a match. Oh, and there are always disguises.
Boston airport tried the system and has given up after 9 months. There were just too many false positives. So, we'll have to wait and see if the software and hardware become sophisticated enough to do a credible job. Right now - it just isn't there.
Addendum to the inflated claims mention above. If you note in the article mentioned, the company claims a 99% accuracy rating. I'm sure they have some proof of that, but it would be easy for any company to make a claim. If you aren't technical enough you could be taken in with the workings of gadgetry and miss the fact that the system doesn't exactly meet it's stated claims.