Computing Applications

Big Brother Is Watching You

Showing the aspects of the face considered in facial recognition.
Facial recognition technologies are increasingly being used by government agencies and commercial enterprises, despite privacy concerns.

Imagine you are walking across a city park and you accidentally drop a gum wrapper. When you return to your office 10 minutes later, there is an email from the local police department informing you that you have been seen littering, identified, and issued a ticket; you can avoid a court appearance by paying your $50 fine by mail.

Outraged? Don't be. The technology is ready, and it could happen to you.

Chinese state news media recently reported that systems of government-owned surveillance cameras, face-image databases, and facial recognition software were nabbing jaywalkers by the thousands. Within minutes of being filmed crossing a street against a red light, the jaywalker's face and identification, including home address, are posted on a public sign. The law-breaker is shamed by the public posting, and also must pay a fine or perform community service.

Of course in the U.S., the arm of the law is not quite so long, and citizens' demands for privacy are stronger. However, evidence suggests that anything that technology can do, and the law does not forbid, will be done. Law-enforcement and commercial interests will see to that, says cryptography pioneer and privacy advocate Whitfield Diffie. "It seems to me another battle lost in the relationship between people and
 machines," he says. "I'm impressed that facial recognition has become so cheap that it can be applied to catching jaywalkers."

Diffie, recipient (with Martin Hellman) of ACM's A.M. Turing Award for 2015 for "fundamental contributions to modern cryptography," may be impressed, but he's hardly pleased. "There was a natural limitation on limitation of freedom in a society where people had to do all the law enforcement," he says. "There were only so many police, and they could only enforce a certain amount of the law. As it becomes acceptable for machines to perform law-enforcement functions, we will get more and more of it. We will be watched for more and more trivial transgressions."

That is deeply troubling, says Anita L. Allen, a professor of law and philosophy at the University of Pennsylvania. "People say that what happens in your home is private, but once you leave your house, everything you do is public and you have no expectation of privacy," she says. "But I have never accepted the idea that privacy interests end at your front door. In fact, a lot of people leave their homes in order to get more privacy."

Allen says privacy law in the U.S., such as the Privacy Act of 1974 (which governs the collection, maintenance, use, and dissemination of personally identifiable information about individuals in federal records systems), has not kept up with advances in technology. In any case, the law is not completely clear. "We have a lot of judges who accept the position that if it happens outside your house, it's 'open season'."

While Allen says she has no problem with systems that detect and identify people in some public spaces, such as in front of ATM machinesf or perhaps in certain high-crime areas, she said there ought to be public spaces from which cameras are forbidden. Determining which spaces are thus deemed private would be the joint responsibility of law enforcers and city planners, she says.

It's not surprising that countries with little or no tradition of privacy rights, and with governments that put their own interests above those of individual citizens, have in recent years led the way in developing surveillance systems based on powerful computer technology. Deva Ramanan, an associate professor of robotics at Carnegie Mellon University, says, "In the past five years, there has been a considerable increase in research originating in China, all throughout computer vision, especially face recognition and face analysis."

Hangzhou, China-based Hangzhou Hikvision Digital Technology Co. is the world's leading provider of video surveillance products and systems. A recent press release from the company said its technologies can "replace the large amount of manpower previously required for searching surveillance footage, detecting anomalous data, and finding ever more efficient ways to allow surveillance to shift from post-incident tracing to alerts during incidents—or even pre-incident alerts." Hikvision says it will soon introduce products "with deep learning technology, such as the DeepInview Series cameras which can accurately detect, recognize, and analyze human, vehicle, and object features and behavior." Another of its products uses "advanced deep learning algorithms [to] imitate human thoughts and memory."

The increasing power of such AI-enabled systems raises privacy issues, but concerns run deeper. Even if a piece of information, such as a digital photograph from a public place, has been legally obtained, some fear it could be improperly shared with or sold to third parties.

China's Hikvision sells its products in more than 100 countries, including the U.S. Last year, the U.S.-funded Voice of America (VoA) News posted an article suggesting Hikvision systems are sending surveillance data secretly back to the Chinese government: "Imagine a world where almost everyone can be tracked, and everything can be seen by cameras linked directly to the Chinese government. The rapid growth of a little-known Chinese manufacturer of high-powered surveillance technology has some people concerned that it's no longer a theory." VoA offered no proof this was happening, and Hikvision denied its technology is used for such purposes.

Social and legal issues aside, rapid advancement in computer vision technology will ensure that facial recognition will continue to become cheaper, more accurate, and more pervasive. Ramanan says a "perfect storm" of factors underlies these trends: faster and cheaper computer processing based on graphics processing units (GPUs), huge databases of labeled data to train artificial intelligence engines, and better and cheaper digital cameras.

Improvements also are being made in the deep learning algorithms behind facial recognition. For example, Ramanan and robotics Ph. D. student Peiyun Hu have developed algorithms that extract and analyze the context that surrounds a face,  such as hair, clothing, or movement. While that may not directly enable the face to be identified, Ramanan says, it can be quite useful in detecting the presence of a face, which is the first step in facial recognition.

Gary Anthes is a technology writer and editor based in Arlington, VA.

Join the Discussion (0)

Become a Member or Sign In to Post a Comment

The Latest from CACM

Shape the Future of Computing

ACM encourages its members to take a direct hand in shaping the future of the association. There are more ways than ever to get involved.

Get Involved

Communications of the ACM (CACM) is now a fully Open Access publication.

By opening CACM to the world, we hope to increase engagement among the broader computer science community and encourage non-members to discover the rich resources ACM has to offer.

Learn More