Most of us have seen the scene in Minority Report in which Tom Cruise is addressed personally by billboards that recognise him as he walks by. For techies it's cool; for the rest of us, a little creepy.
Once the stuff of science fiction, facial-recognition software has been in use for more than 10 years. It was deployed at a US airport after the 2001 terrorist attacks, having been used to scan everyone entering the stadium at the Super Bowl earlier in the year.
Now, it's gone mobile; a company in Massachusetts has developed a device to attach to police officers' iPhones that lets them identify a person from the police database simply by taking a snap.
Every day, 100m photos of people are tagged with names on Facebook. In an attempt to accelerate this, the social network recently caused a storm by turning on facial-recognition technology, automatically suggesting names. Google's Picasa has had the technology for some time, and Apple recently acquired Polar Rose, a specialist tech company, to add this service to iPhoto.
It is even in use (with varying degrees of success) in 'fun' applications. Genealogy site MyHeritage.com has a page that compares a picture of you with its celebrity database to determine whom you most closely resemble; Jessica Lange, by 66% if you're asking. It's better than Google Goggles, however, which thought I most closely resembled a photo it held of a pink milkshake.
So the Feds have it, as does Madison Avenue. It is a recent study by researchers at Carnegie Mellon University that gives far greater cause for concern, however, drawing on publicly held information to identify and reveal personal details of people photographed in the street.
The researchers conducted three experiments. In the first, they took anonymous profile pictures from a popular dating site and were able to put names to them using off-the-shelf software to compare these with public Facebook profiles, without even logging into Facebook.
The second was about offline-to-online identification. Taking pictures of students on campus and scanning them against Facebook, they were able to identify about a third of subjects.
Finally, having photographed and identified people, they were able to infer personal data about them using cloud computing and data mining to seek out sensitive information like Social Security numbers.
What Facebook and other services are doing is pretty harmless in itself; simply saying 'here's that person again' and helping to organise your photos. But the study shows how straightforward it is to turn this arguably innocuous activity into a privacy hornets' nest.
The research demonstrates that it isn't just Facebook we should worry about when we put our photos on its site. The social network forms a crucial part of our digital footprint, its insistence on using real identities making it the connecting point from a face to a name, from which a wide variety of information can then be sourced and inferred.
Earlier this year I wrote about The Onion's spoof claim that Facebook had proved a fantastic investment for the security services as, across the web, people chose to record their actions, location and relationships - saving the spooks millions in snooping costs.
What's becoming clear, whether you subscribe to the conspiracy theory that Facebook is just a giant Homeland Security database or not, is that it is part of a web of sites that, intentionally or otherwise, are contributing to the death of anonymity; not just online, but as you walk down the street, too.
Andrew Walmsley is a digital pluralist
30 SECONDS ON ... FACIAL-RECOGNITION SOFTWARE
- Automated facial recognition has proved difficult to get right. The two main methods are geometric - looking at distinguishing features - and the statistical photometric approach.
- Many traditional 2D systems use 'landmarks' on the face, such as the size, shape or position of the eyes and nose, as points of comparison.
- Other 2D systems may use face data derived from a normalised gallery of images for comparison.
- In recent years, 3D facial-recognition has become a practical option. Sensors allow the capture of more detailed information about distinctive features, including contours. This method is less likely to be thrown off by differing lighting and angles. The addition of skin-texture analysis reportedly improved facial-recognition by up to 25%, according to research at the Carnegie Mellon Robotics Institute in 2008.
- The US Department of State has a system containing more than 75m images that it uses for visa processing. Australian Customs also uses facial-recognition technology.
- At Super Bowl XXXV in January 2001, Viisage software was used by Florida police to identify potential criminals in attendance. It reportedly matched 19 faces from the 100,000 scanned, but none of them was among those ultimately arrested that day.