42-15483605 fmt

MAY 2013

Three community members discuss the role of facial recognition technology in their everyday world.


Facial recognition technology is an important, proven tool that has gotten better over the years with new iterations. Just over a year ago, New Jersey undertook a project called “Operation Facial Scrub,” which utilized facial recognition technology to “scrub” its 19 million photo record database. Of the potential photo matches that resulted, nearly 4,900 matters required administrative corrections to the database. This is important because that large number could have created problems for customers…this eliminates customer inconvenience.

The scrub also resulted in the discovery of more than 2,600 separate acts of criminal fraud against the Motor Vehicle Commission—people attempting to change a record or adopt an alternate identity. To date, we have sent 750 criminal cases to the Office of the Attorney General or other federal, state and local law enforcement agencies for investigation and/or prosecution, with that number expected to grow. This is all thanks to facial recognition technology, because there’s no way we would have found these individuals through our normal course of business.

In addition to security, there is also a customer service benefit. New Jersey was one of only a few states that required everyone to come into the office every time to renew a driver’s license. After cleaning up the database, we have been able to invite select customers to renew their driver’s license through the mail through our “Skip the Trip: Renewal by Mail” initiative. We are able to do this, as well as plan to expand the initiative, because we are now even more confident that the record we have on file for an individual is correct because it has been vetted through the facial recognition software.

Like many things, facial recognition (FR) is a tool—a tool to help us detect fraud. It is not a silver bullet, nor is it fool-proof. But it can be a significant aid to finding those who for nefarious reasons seek to acquire multiple valid driver’s licenses. The technology has proven its merit in a number of states, with both statistical and anecdotal evidence of its utility. Over the past 10 years, FR performance has continued to improve—both for constrained environments as well as unconstrained. Outside of the traditional DMV applications, we’re seeing FR used online by the major search engines and mobile device capture supporting a number of applications including online user authentication. Could this open new applications in the DMV arena as well? 

For example, could DMV’s offer more services online if we could be more confident in the identity of the person on the other end of the network?  Beyond passwords and knowledge based authentication, facial recognition could be used for this purpose—capturing the facial image using the camera embedded in a laptop or mobile device and comparing it to the driver’s license photo—providing multifactor authentication. Just a possibility.

Face recognition is a very useful fraud detection tool for administrators issuing licenses and IDs. Today, over 40 license jurisdictions in the U.S. use face recognition to check for duplicate applications, “scrub” existing records to flag duplicates, and/or to verify renewal identity.

While face recognition algorithms are still not as accurate as fingerprint identification, face recognition performance has improved dramatically over the last few years according to results from independent government testing.

Administrators can make face recognition an effective tool in their fraud prevention arsenal by following these best practices:

  • Collect facial images that conforms to recognized standards to enhance matching performance.
  • Implement software capable of flagging defective images at capture.
  • Capture at a resolution of around 120 pixels between the pupils.
  • Choose high quality cameras with good lighting and minimal distortion.
  • Train operators to collect images that exhibit proper pose and expression.
  • Use compression techniques that do not impede accurate feature. detection and matching (target fie size should be no less than 40KB)
  • Ensure that you have adequate staff to adjudicate match candidates.