|
WASHINGTON — Homeland Security leaders are
exploring futuristic and possibly privacy-invading technology aimed at
finding terrorists and criminals by using digital surveillance photos that
analyze facial characteristics.
The government is paying for some of the most
advanced research into controversial face-recognition technology, which
converts photos into numerical sequences that can be instantly compared
with millions of photos in a database.
Facial-recognition research was sought to enable
federal air marshals to surreptitiously photograph people in airports and
bus and train stations to check whether they were on terrorist databases.
The air marshals disavowed the technology to focus on identifying suspects
through methods that don't use cameras.
Even so, the research continues and could help police
identify someone photographed by a security camera, said David Boyd, head
of command, control and interoperability at Homeland Security's Science
and Technology branch.
The technology has been tested at Boston's Logan
International Airport, in busy city centers and at the 2001 Super Bowl in
Tampa
The ability to establish quick identities will
"turbocharge video surveillance," ACLU privacy expert Jay Stanley warns.
"It turns 'dumb' camera lenses into 'smart' observers that not only
capture images but attach an identity to the image. That could increase
the attractiveness of surveillance cameras."
Melissa Ngo of the Electronic Privacy Information
Center says the technology endangers privacy by enabling ordinary security
cameras to find out the names of people being observed. "Why are you being
tracked if you're not doing anything wrong?" she said.
Face-recognition cameras have helped casinos spot
known card counters and other unwelcome gamblers, said Walter Hamilton,
chairman of the International Biometric Industry Association.
More recently, 19 states have adopted the technology
and compare driver's-license applicants with a photo database of license
holders to see whether an applicant already has a license or is using a
false identity, Hamilton said.
The Homeland Security research aims to make the
technology work in one area where it has failed: surveillance. Tampa and
Virginia Beach police removed face-recognition systems that did not yield
a single arrest. During a test at Boston's airport in 2002, the system
failed 39% of the time to identify volunteers posing as terrorists at
security checkpoints.
Using face-recognition for surveillance is
"enormously difficult" because systems photograph people at oblique angles
or in weak light, both of which create poor images, said Takeo Kanade,
head of Carnegie Mellon University's Robotics Institute. Terrorists can
defeat the systems with disguises or hats that shield their faces.
The Homeland Security research aims to counter
shortcomings by creating technology that will "take a partial picture of a
face and reconstruct that into a full frontal shot," Boyd said. "No one
has done that before."
Kanade said the research, by L-1 Identity Solutions
of Stamford, Conn., "challenges the most difficult part of face
recognition. It's a challenge worth
pursuing." | 
CONTRACT
DETAILS
WASHINGTON — Homeland Security leaders are
exploring futuristic and possibly privacy-invading technology aimed at
finding terrorists and criminals by using digital surveillance photos that
analyze facial characteristics.
Facial-recognition research was sought to enable
federal air marshals to surreptitiously photograph people in airports and
bus and train stations to check whether they were on terrorist databases.
The air marshals disavowed the technology to focus on identifying suspects
through methods that don't use cameras.
The technology has been tested at Boston's Logan
International Airport, in busy city centers and at the 2001 Super Bowl in
Tampa
