By Deepa Seetharaman
Facebook Inc. says it has developed a way to recognize people in
photos even if their faces are obscured.
"We can easily picture Charlie Chaplin's mustache, hat and cane
or Oprah Winfrey's curly volume hair," Facebook said in a paper
presented earlier this month. "Yet, examples like these are beyond
the capabilities of even the most advanced face recognizers."
The findings, first reported in New Scientist, a weekly
technology and science magazine, stem from Facebook's work in
computer vision, which involves teaching software to recognize
language and images as quickly and easily as the human brain. The
findings themselves aren't considered a big breakthrough but
represent the latest step in an emerging body of work from
Facebook, experts said.
In the paper, researchers at Facebook and the University of
California, Berkeley used body shape, posture and other clues to
identify people in photographs even if their faces weren't entirely
clear. Facebook said it can identify individuals with 83% accuracy
using a method dubbed PIPER, an acronym for pose invariant person
recognition.
Facebook tested PIPER on more than 37,000 photos from public
Flickr albums with more than 63,000 images of people. Slightly more
than half the photos contained images of faces at a resolution
suitable for recognition. Researchers focused on people who
appeared at least twice in each album.
"Machines can understand people in images much better than five
years ago and in some cases as well as people can," said Nigel
Duffy, chief technology officer of the San Francisco-based
artificial intelligence startup Sentient.
The method also allows Facebook to identify someone partly
through their social circles, said Aditya Khosla, a researcher at
MIT's Computer Science and Artificial Intelligence Lab.
The technology could help Facebook develop more products akin to
its newly launched photo-sharing app Moments, which uses face
recognition to group images based on who is in each photo.
In the future, this technology could help advertisers,
law-enforcement officials and academics mine photos for clues about
a person's activities, interests and social circles, experts
said.
Those capabilities also may draw scrutiny from regulators and
privacy advocates in Europe and the U.S.
"If you have people's faces recognized without their consent,
you're violating their right to a private life," said Mike Weston,
chief executive of London-based data-science consulting group
Profusion.
Mr. Khosla said Facebook could mitigate concerns by letting
users limit when they can be tagged in photographs.
Facebook didn't respond to requests for comment.
Facebook has been investing heavily in studying artificial
intelligence since late 2013 when it hired well-known French-born
researcher Yann LeCun.
Facial-recognition technology, also known as "faceprint"
technology, is a particularly complex problem, Mr. LeCun has said.
Clothing can help identify someone, but it isn't always a good
indicator, researchers said, because people change clothes or
multiple people may wear the same outfit.
Other tech companies and government agencies use facial
technology as well, including the U.S. Department of Homeland
Security and the FBI. Google Inc. last month introduced an app,
Google Photos, that uses facial-recognition technology to identify
people in photos.
The app is available world-wide, but the facial-recognition
features are available only in the U.S. Likewise, Facebook won't
offer its Moments app immediately in Europe because of regulators'
concerns about its facial-recognition technology.
Access Investor Kit for Facebook, Inc.
Visit
http://www.companyspotlight.com/partner?cp_code=P479&isin=US30303M1027
Subscribe to WSJ: http://online.wsj.com?mod=djnwires