Indiana motor vehicle agency uses facial recognition technology to deter identity theft.
The Indiana Bureau of Motor Vehicles (BMV) has taken another step toward warding off identity theft by deploying facial-recognition technology in all its 140 branches.
In 2007, the BMV began issuing digitally produced drivers' licenses with bar codes and holograms. Then the BMV joined most states by connecting to the Social Security Administration to check Social Security numbers online.
Now, the BMV joins more than 20 states by adding facial-recognition features to new licenses. The BMV launched a pilot project in November 2008 in three branches to develop procedures, then went live statewide only a month later.
"We felt it was incumbent upon us to accomplish our goal, which is to affirm to our public that their driver's license is accurate and secure," said Dennis Rosebrough, Indiana BMV communications director.
A stolen license has much more value on the street than it did previously, Rosebrough said. "Ten years ago, a fake driver's license might have gotten you a six-pack, whereas today the government-issued picture ID obviously has higher stakes relating to cashing checks or getting on a plane," he said. "Its value has been ratcheted up. In that environment, we're even more compelled to maintain the security and accuracy of our data and credentials."
When a citizen applies for a new license or renews an existing one, his or her photo is taken and "enrolled" in the system. Each evening, the BMV's system runs the photo against existing images in a database of approximately 6.5 million license holders. Using algorithms, the system produces a score that indicates the probability of a match with any existing photos.
Scores over a certain threshold are reviewed the next morning by personnel. Throughout Indiana's 140 motor vehicles branches, there are usually about 500 reviews daily, or about 5 percent of the images processed -- most of which are attributable to similarities between two people or data that was manually entered wrong. Each day, the system usually uncovers two to three suspect cases, which are handled by an internal security and investigations unit.
It's taken time to develop such an efficient solution. States began deploying facial recognition about eight years ago, and the technology has improved over that time. "The algorithms are cleaner, faster, more reliable, and the cameras have improved. With more megapixels, you're getting more detail at the front end of the capture piece," said Dino Redmond, regional director of secure credentialing for L-1 Identity Solutions.
"You're not trying to pick a face out of the crowd," Redmond said. "You have a subject 60 inches away from the camera, and they have appropriate lighting and a consistent background. It's absolutely ideal from a biometric standpoint."
The software relies on a clean image, so applicants must remove hats and glasses and not smile too broadly when having their photos taken. "We want the system to operate as efficiently as possible, and anything that's a distortion of the person's face limits the accuracy of the reference points," Rosebrough explained. "We really emphasize that the technology sees the permanent facial features of points of measure, like distance between the peaks of the cheekbone or the distance between the eyes."
The software creates a digital map of the face and collecting data gleaned from a face's features. It then creates a template -- a stream of bits and bytes that identifies the uniqueness of each image. It takes just seconds to scan the existing database of photographs and return images with which there's a high probability of a match.
The software could be a boon to law enforcement, because it quickly matches photos with names. "One of the ways the system has advanced over the years is it allows portfolios
to be created," Redmond said. "You could set up a portfolio that has 'most wanted' so that immediate searches could be done."
The BMV entered into a contract with L-1 Identity Solutions in August 2008, and the project was finished just four months later. The system is basically an off-the-shelf product that's deployed and maintained by L-1 through the life of the four-year, $2.4 million contract.
"It's our responsibility to provide the software and drive the integration process," Redmond said. "It's our own servers and technology. "We deliver the goods, they host it and we maintain it over the life of the contract."
One of the first tasks for L-1 was to "cleanse" Indiana's database of duplicate photos. The database contained about 13 million images dating back to 1999. In some cases, there were multiple photos for a single driver's license number because of license renewals; those images had to be matched with a single license holder. The database was pared down to about 6.5 million.
Other than that task and minimal training for BMV personnel, the deployment was quick and easy. "We've been able to create a repeatable process and make it efficient and cost-effective," Redmond said.
"It's not like we were groundbreaking with the technology," Rosebrough said. "But we needed to make sure, from an implementation standpoint, that we had the operational procedures for our associates in the branches, as well as the internal computer programs and integration of the technology with our platform. That's why we did the pilot."
The BMV expects to add about 1.8 million licenses yearly to the system and predicts it will be searching through 14 million every day by the end of the contract. At the end of the four-year contract, the state will own the system or it can renew the contract. The state would then own the hardware and images. If Indiana renews the contract, L-1 would modify the system routinely with the latest technology.