The human face does more than help us tell each other apart. Studies have suggested that our perceptions of attractiveness and personality traits — even intangible qualities such as trustworthiness and sincerity — could be based on instinctive responses to certain facial features.
But recent technological breakthroughs could soon lead to a day when computers know us better than we know ourselves.
Facial recognition technology has been used in recent years to screen for criminals, to catch cheaters at casinos, and even to recommend products at vending machines. With a little more refinement, it could eventually surpass ID cards and fingerprints as the new standard for identification. In the U.S., a country with more than 30 million closed-circuit surveillance cameras, facial recognition could also be used to track individual people as they go about their daily lives.
These are just some of the latest advancements in a growing pack of products that incorporate digital human face recognition, and countless others are being developed across a spectrum of industries including medicine, computing, gaming, law enforcement and advertising.
Dr. Shalini Gupta of AT&T Laboratories, who earned master’s and doctoral degrees in Electrical and Computer Engineering from The University of Texas at Austin in 2004 and 2008, worked with a team to develop algorithms for 3D human face recognition in a project supervised by the university’s Biomedical Informatics Lab and Laboratory for Image and Video Engineering.
The team mapped and measured the facial features of 118 subjects, recording proportions — such as the distance between eye corners and the width of the nose — that make each 3D model unique. The algorithms they created outperformed several of the standard face recognition methods that are currently in use.
As facial recognition generates buzz in scientific circles, the technology is also making an impact in the business world, Gupta says. She outlined the future of the field recently at The Austin Forum on Science, Technology and Society.
“All of the big tech companies are investing R&D dollars in this area,” she says. “Google, Microsoft — everyone is looking at it. It’s a very hot topic. It’s not going to be long until we see face recognition on cell phones.”
Google made a foray into facial recognition when it acquired Neven Vision — a company that pioneered early visual search and analysis technologies — in 2006. Its Picasa program includes a feature that scans online photo albums and automatically identifies people who appear in multiple images and sorts the photos accordingly.
Apple’s iPhoto and Microsoft’s Windows Live Photo Gallery have similar tools, and in December, Facebook announced a new feature called Tag Suggestions that will identify friends’ faces in uploaded photos so they can be tagged more efficiently.
These 2D systems became available to consumers in the last decade or so, but Gupta says many companies have already begun to shift toward developing 3D and mobile face recognition technologies.
Apple appears to be making a push to integrate face recognition into mobile devices — possibly the next wave of iPhones and iPads — evidenced by its acquisition of the face recognition company Polar Rose last fall.
Once perfected, 3D systems could be even more useful in identifying people in real-world environments because they account for variations in facial structure and pose. Still, there are a number of hurdles that must be overcome before biometrics technology achieves its full potential.
“I don’t think the facial recognition problem is solved. Specifically, it is not solved for unconstrained environments where we can’t control backgrounds, lighting conditions, and poses,” Gupta says. “These are all things at need to be resolved before we can get to a point of high accuracy.”
Some medical institutions, including the Dell Children’s Hospital in Austin, have already begun to use 3D facial imaging systems to take measurements and screen patients before surgery.
Security and Privacy
Face recognition is also being refined as a tool in security and law enforcement. The technology differentiates itself from other identification methods because it can be used covertly, and unlike ID cards or pin numbers, faces are unique and cannot be lost or stolen.
There are, however, a number of practical and social limitations to its use in surveillance: The technology is more accurate when used to verify an individual person’s identity than it is when used to identify unknown subjects lurking among large crowds. The potential for false positives makes this method less than ideal in uncontrolled environments, at least for now, Gupta said.
These issues were brought to light following the controversial use of a scanning system at the 2001 Super Bowl in Tampa, Fla. Scanners photographed attendees without their knowledge as they entered the stadium, and the system flagged 19 people that reportedly had outstanding criminal warrants. Public backlash over the secret recordings led officials to abandon the program the following year.
Another limitation of current facial recognition technology is that it tends to have trouble identifying people who attempt to fool the system by wearing disguises, but Gupta says that problem is solvable.
“A lot of the work that has been done in the research has been in a controlled setting, which means you have cooperative subjects,” she says. “There’s a certain degree of robustness built into the system — but there’s not a whole lot, to be honest. Going forward, that would be a big part of what would need to be solved — to actually identify people that are trying to fool it.
“There are things people can change about themselves — hairstyles, hats, scarves, etc. — but there are certain physical features you can’t change, like the distance between your eyes,” she said.
In an infamous scene in Steven Spielberg’s 2002 film Minority Report, a network of cameras at a shopping mall scans the eyes of shoppers as they move through a crowd, allowing each of them to be bombarded with a flurry of customized holographic advertisements that address each customer by name.
While the system portrayed in the film is fictional, the technology behind it might be just around the corner. Last year, Northeastern University professor Russell Pensyl developed a prototype of “an unobtrusive, ubiquitous content delivery system” that uses facial recognition to estimate a person’s age and gender and then recommends products of interest to their specific demographic.
Gupta says there are numerous opportunities for the technology that might have seemed impossible just a few years ago.
“When I started working in this area, I didn’t realize how real it was,” she says. “It’s out there. People are working on it and developing products. It seems like science fiction, but it’s real.”