Facial-recognition technology needs limits, privacy advocates warn

Consider, for a moment, the amount of information that’s contained in your face. It’s a gold mine, a collection of identifiers unique to you, from its proportions to its features to the way it moves. A quick look tells your life story, indicating your sex, racial background and age, and, in the moment, your attentiveness and mood. Humans come hard-wired to intuit this information. But now, computers are catching up and learning to make sense of all manner of facial information gleaned from photos and videos.

Before long, we’ll be living in a world in which store signs know what kind of customer is looking at them, where home-security cameras know who’s coming through the door and when Facebook will be able to tell if you’re in a photo the moment it’s uploaded. The distant future? Try next month. Your face is about to become big business and that has some players excited and others worried that citizens’ control over their own privacy will be gravely eroded.

Facial-recognition technology has long been the domain of security agencies, which have used it to verify official documents, help secure borders and assist in policing duties. And while this field is only growing, the explosion of ever-faster computers, ever-cheaper cameras and growing network capacity has seen the technology making inroads into the consumer world as well.

In its rudimentary forms, consumers are already familiar with it: For years, point-and-shoot cameras have been able to pick out the faces in a scene and adjust the shot accordingly. Webcam software has long been able to spot faces, if only to overlay a mariachi hat and mustache in the right spot. But spotting faces and putting names to them are two different things.

Two key elements are required for facial recognition: Software that analyzes images and databases of identifying information that can link faces to names. One isn’t much good without the other – but the Internet is bringing them together.

One online behemoth is uniquely poised to put a face to a name.

Facebook, a company with more than 750 million users and a spotty privacy record, is rolling out facial-recognition technology that will automatically suggest names of friends in photos that are being uploaded. Already available in many countries – but not yet in Canada – the feature is kicking up a cloud of protest.

The feature only identifies people on the user’s friends’ list, so you won’t be tagged in the background of a stranger’s Eiffel Tower tourist snapshot. People who upload photos sign off on the identifications; users who are identified are notified and can un-tag themselves. However, like many Facebook innovations, the feature will be activated by default – users will have to opt out if they don’t want their faces scanned.

In the United States, a coalition of prominent digital-rights groups has filed a complaint with the Federal Trade Commission, alleging, among other things, that users aren’t fully informed of the biometric information that’s “secretly” being collected about them. They also fear that Facebook might be more willing to share this data with eager advertisers and less-than-scrupulous app developers.

Facebook has not announced if or when its feature will roll out in Canada, but privacy advocates are worried here, too.

“I think it’s appalling,” says Ann Cavoukian, the Information and Privacy Commissioner of Ontario. “It is very sensitive information, and to rely on ‘Trust us, we’re Facebook’ – they’re asking too much.”
Ms. Cavoukian, who has done extensive work internationally on ensuring that information collected about users cannot be used against them, says that privacy concerns arise when large databases of biometric identifiers – things like faces and fingerprints – can be matched against identifying information, like names.

Facebook not only has access to a vast store of names and pictures of faces – which it’s moving to connect faster than ever, by using facial recognition to facilitate the tagging process – those faces are connected with places and events, that, when combined, could create a sweeping picture of who was doing what, where and when.

“You could have gone to a party and not thought twice about it, and maybe there’s an indiscriminate photo taken,” says Ms. Cavoukian. “It could be automatically tagged. Then they found out something happened at the party. The police come a-calling and you get a knock on the door.”

The public might be accepting of this scenario in the case of a murder. But what if the police decided to crack down on marijuana use? Or untangle associations of political opponents? Canadians might not have to worry about such implications today, but Facebook users under more authoritarian regimes might not be so lucky.

Facebook declined to be interviewed about the technology, though a public-relations representative explained that the company doesn’t like the term “facial recognition” – it prefers “Tag Suggestions” – and considers the feature to be “opt-in,” since it only works on photos that have been newly uploaded and users can disable the function should they not wish to be identified.

The Privacy Commissioner of Canada’s office – whose previous tangles with Facebook prompted the company to improve its privacy practices – has also confirmed its interest, though it’s not actively investigating at the moment.

So far, other online giants have been more circumspect. Apple and Google – two companies with access to vast quantities of user data – have both developed or acquired facial-recognition technology, but so far have only applied it to software, like iPhoto or Picasa, that scans users’ private photo albums, not the open Internet.

Google actually built a working facial-recognition search engine into its “Google Goggles” mobile phone app, which would let users take a photo of someone’s face and use Google’s vast database to match it to other information. However, the company decided not to release it, citing privacy concerns.
“People could use this stuff in a very, very bad way as well as in a good way,” Google chairman Eric Schmidt told a conference last month.

View to a sale

Facial detection and recognition are hardly confined to the Internet. These technologies are also moving offline, into your home, workplace and neighbourhood.

Computer-vision technology is already critical to iWatchLife. The Ottawa-based company sells smart surveillance systems: Not only do cameras keep an eye on the home, but computers pay attention to what’s going on in the scene, alerting homeowners should it recognize certain events.

“My wife and I went away for the weekend,” says Charles Black, the president and CEO of iWatchLife. “I’ve got two teenage daughters. I told my system to let me know if more than ten people walk into the house.”

Now, the company is working on adding a facial-recognition component to its products. It could work as an access-control product for small businesses – allowing only certain people into a stockroom, for instance. And it could be used by security-conscious families, too.

“You don’t want to be Big Brother, but at the same time, if your kid is coming home from school, you want to make sure they’re home safe,” says Robert Laganière, an engineering professor at the University of Ottawa and iWatchLife’s chief scientist.

Your front door is just the beginning. Face-sensitive cameras will soon be populating malls and upscale retail outlets, picking out customers and tailoring displays to suit them. Intel, the chip-making giant, is working on facial-detection systems that don’t identify customers so much as profile them.

Using software developed by Markham, Ont.-based CognoVision (which Intel acquired late last year), a camera mounted on a large LCD screen watches for faces that come within four to six metres. The screen can switch ads depending on the kind of face that walks by. The captured images are deleted and the individual customers are never identified – a system that Ms. Cavoukian lauds for ensuring privacy by its design.

In addition to boosting sales, the system also collects information for retailers. By measuring who’s looking at the screen, and for how long, the system can gauge the effectiveness of different pitches. “For a retailer, that’s incredibly powerful information,” says Christopher O’Malley, director of retail marketing with Intel’s Embedded and Communications Group. The technology has already been piloted and Intel says it plans to announce retail deployments soon.

Facial-recognition technology isn’t perfect yet. Academics say it still struggles with low-resolution or grainy photographs and can still be thrown off by facial changes like glasses or a beard. Cheap facial-recognition systems can even be fooled by holding a picture up to the camera.

But algorithms are getting smarter, learning, for instance, to stitch together multiple frames of grainy video-camera footage into a recognizable image. What’s more, “learning” software that trains itself to recognize a face has a huge advantage if it can draw on a big database of names and faces to use as a reference. Once upon a time, only institutions like governments and insurance agencies maintained such files.

Times, though, have changed. Processing power is cheap, cameras are everywhere and now facial-recognition software has access to a certain book of 750 million faces to learn from. Facial recognition isn’t just for Big Brother any more; it’s for little brother too. Say cheese.

Origin
Source: Globe & Mail