Facial recognition

On June 30, The Dickinson Press website posted Robb Jeffries’ article on the use of facial recognition by several state governments. Certain states, including North Dakota and Minnesota, use photos from driver’s licenses and other identification cards to compile databases that can be scanned with facial recognition software. Both states use the images to detect identification fraud.

Speaking for the Minnesota Department of Public Safety, Doug Neville said that in 2008 his agency searched about 11 million identification records to find duplicate photos. Their search flagged almost 1.3 million identification records. A review of the records revealed that many of these simply indicated people who had changed their name or twins. Nevertheless, the project led to the cancellation of nearly 10,000 identification cards.

ACLU senior policy analyst Jay Stanley stressed that a cautious and transparent use of facial recognition technology is vital for widespread acceptance. “When it comes to investigating specific crimes (with facial recognition technology), there needs to be checks and balances,” Stanley said. “This is something we see top to bottom in law enforcement, where they are making very profound decisions about what they do with this technology without input from the American public.”

Although facial recognition technology is often presented as a method approaching fingerprint ID in accuracy, this is not yet true. Facial expressions, photo angles and photo quality affect the outcome of facial recognition analysis.

Glenn Jackson, driver’s license director with the North Dakota Department of Transportation, told Jeffries that his agency’s software works best with head-on photos. “The system we have is very basic,” Jackson said. “It’s not like that ‘NCIS’ stuff where you can bring up half a cheek and know who it is.”

Facial recognition. Source: The FBI’s Biometric Center of Excellence.

On TV shows and in movies, computer experts start with a blurred image of a suspect. They very quickly “clean up” the image to generate a clear photo and get results from facial recognition software. In real life, the facial recognition process does not work that way.

During May, Sean Gallagher posted an overview of the technology on the Ars Technica website. Entitled, “Why facial recognition tech failed in the Boston bombing manhunt,” Gallagher’s article surveys the procedures and pitfalls of this technique.

The first step of facial recognition requires the detection and enhancement of an image of a face. This is not a quick and simple procedure.

“Hollywood does a pretty good job of creating a myth that you could extract a better image by enhancing and zooming where information wasn’t captured,” Masayuki Karahashi told Gallagher. “You’re not going to create more information out of nothing.” Karahashi is 3VR’s senior vice president of engineering for surveillance and video analysis technology.

In one approach to the second step, software identifies common facial features, such as the centers of eyes, to produce a faceprint, which is a mathematical representation of facial patterns. During the final step of a facial recognition search, the faceprint is checked against a database of faceprints.

As indicated by the title of his article, Gallagher looks closely at the use of facial recognition technology during the hunt for the Boston bombers. The exercise highlights challenges of employing a sophisticated technique with images captured from uncontrolled environments.

As an aside, Gallagher mentions an interesting bit of history: Las Vegas casinos numbered among the first organizations to use facial recognition to keep card counters and other unwelcome guests from the gambling floors.