Face recognition useless for crowd surveillance
So why is Visionics selling it to airports?
Anyone offended by the Orwellian implications of using face-recognition technology to scan airport crowds for terrorists can take heart in the fact that the technology is, quite simply, worthless in that situation.
As an authentication tool, used in controlled settings, face recognition has real value. But even here we can expect a false acceptance rate (FAR) of one in 250, according to biometrics outfit FaceKey.
"This means that under controlled circumstances....you could expect one false positive out of 250 people when face recognition is used alone," FaceKey COO Annette Starkweather told The Register. "FaceKey has combined face recognition with fingerprint recognition to [achieve] a FAR of one in 2.5 million," she added.
"Limiting access to secure areas in airports would be a perfect application for biometrics," Starkweather says.
But in uncontrolled settings, such as we'd encounter in a surveillance context, the performance of face recognition falls to absurd depths.
This has actually been examined by the US Department of Defense (DoD) Defense Advanced Research Projects Agency (DARPA), which sponsored the Facial Recognition Vendor Test (FRVT) 2000, the biggest and most well-known test to date, Image Metrics COO Gareth Edwards told us.
"With indoor light, and a prior image taken at 1.5m camera-subject separations and another taken at 2m camera-subject separations, the best false detection rate (FDR) was 33 per cent, with a false acceptance rate (FAR) of ten per cent."
This means that "to detect 90 per cent of terrorists we'd need to raise an alarm for one in every three people passing through the airport. It's absolutely inconceivable that any security system could be built around this kind of performance," Edwards says.
Apparently, Visionics employs some sort of slick marketing magic by which they run potential patsies through the grease and persuade them to invest in their FaceIt surveillance kit.
"Most worrying is the number of reports from people who've seen working demos and 'field-trials' of these types of systems. Many truly think that they offer an answer. [But] when subject to raw, rigorous analysis, we've yet to see any evidence that these systems offer any value. There's yet to appear any plausible explanation of the results of the FRVT test when compared with so-called 'field trials,'" Edwards says.
The discrepancy, we have to suspect, reflects the natural difference between rigorous testing by disinterested third parties, and some self-serving marketing demo.
Visionics has been sponsoring a public surveillance trial in Tampa, Florida, with the stated goal of busting sex offenders and pedophiles, two target groups which no one would rush to defend. Now they're exploiting the terrorist threat, which in recent weeks has become America's paramount fear.
There is talk that the Department of Transportation will bite the hook and authorize a FaceIt trial at National Airport in Washington, DC when it reopens.
A similar company, Viisage, which made headlines by scanning crowds at last year's Super Bowl, is also eagerly pursuing the airport surveillance angle, and has "offered the FBI free use of their face-recognition technology to aid in the apprehension or identification of the persons responsible for the terrorism in New York City and Washington," for an added marketing gimmick.
The companies are clearly anticipating bundles of cash selling this technology before word gets out that it's of no use in a public surveillance setting.
Afterwards, they can still haul in a nice profit selling incremental 'upgrades' to victims who've invested millions and can't justify backing out; and for an added bonus, they will have become the 'DoubleClicks' of public biometric data, which is sure to be a gold mine in itself. ®
Sponsored: Hyper-scale data management