Facial Recognition

May 2017

WisDMV sees the benefits of using facial recognition technology

Protecting identities and maintaining the integrity of its credentials: These are just two of the reasons facial recognition technology has been a part of the Wisconsin DMV’s business practices since 2005. And with new facial recognition software in place, the jurisdiction is able to catch more instances of fraud, identity theft and criminal activity than ever before.

In September 2015, WisDMV upgraded its facial recognition software to a more robust system. Provided by CBN, the new software includes improved photo equipment in all service centers for clearer facial images, translucent overlay of images for clearer match detection, improved match calibration for more accurate results and digital refinement of all historical photo images to help match new photos to archived photos. What’s more, the facial recognition processes are in real-time, so DMV staff can act much quicker than before.

According to Susan Schilz, the Compliance, Audit and Fraud Unit supervisor for WisDMV, the transition from the old software to the new platform was a bit of an adjustment and a learning process. “We had our parameters set very wide initially, and this robust tool was kicking back far too many images for us to look into,” she explains. “We had to tweak our parameters and deactivate some of our records, such as ones with baby photos. It took us about a month before coming up with a parameter setting that we deemed acceptable.”

But the growing pains were worth the improved results. Today, the WisDMV is flagging more instances of fraud and is finding historical fraud that was not previously caught. That’s because every photo taken each day at the state’s 92 DMV service centers is run through a database of about 21 million photo records.

How it Works

When WisDMV service center employees take a resident’s photo for a credential, they are tasked with performing a 1:1 photo check to compare the new image to the most recent photo on file for that individual. If it matches, they go ahead and process the credential. But if it doesn’t, they take a second look to determine if fraud is being attempted.

“Our service center employees are trained to look at eyes, nose and facial structure,” Schilz says. “But if they’re not sure if the photo should be a match, we [the Compliance, Audit and Fraud Unit] are here to take another look at the image. We can compare it to the other photo records of that individual in our database. This takes the heat off of [the service center staff] to make a decision.”

Taking a Second Look

Images that are not deemed to be a match go into a queue for the Compliance, Audit and Fraud Unit to look at in ‘triage.’ If it still isn’t clear or the team senses fraud, the image goes to analysis to determine whether it’s worthy of an investigation. “This means we’re going to look a little deeper, write letters and determine who is who,” Schilz says.

After an investigation, the unit can take action and require the two individuals to come in and verify who they are. “We’re not always sure who is the victim and who is the suspect,” Schilz says. “They both may need to come in and verify with us.”

Schilz notes that facial recognition is not a perfect science. “There is some skill that’s required, and you do get good at it after a while,” she says. “Our team of eight shares this responsibility so we are all able to hone our facial recognition skills.”

Lending a Hand

Facial recognition technology isn’t just an asset for protecting identities and preventing fraud within the WisDMV. It’s also a tool that can be used to help law enforcement catch suspected criminals or locate missing or abducted individuals.

While law enforcement does not have direct access to the WisDMV facial recognition tool, partnerships have been established to assist in sharing its benefits. “The Wisconsin Department of Justice facilitates photo requests for various law enforcement entities pursuant to Wisconsin statutes,” Schilz explains.

“We collaborate with the Wisconsin DOJ by responding to their requests.”

In 2016, the WisDMV ran 193 photos for law enforcement to help in cases like child exploitation, cybercrimes, missing persons and human trafficking. With an overall success rate of 19 percent, Schilz believes the two hours a week spent on this work is worth her team’s time. “In 30 child enticement cases, we had a 28 percent match and found 11 people that wouldn’t have been found through other means,” she says. “This collaboration has the support of our administrator’s office.”