CMINDS Recognized at International Conference

UMass Lowell Image

Boston Marathon bombing suspects Tamerlan Tsarnaev, right, and his brother, Dzhokhar, are shown together in this photo released by the FBI.

By Edwin L. Aguirre

Three days after twin explosions killed three and injured more than 260 others near the finish line of the Boston Marathon, authorities released photos of the two male bombing suspects, who were later identified as the Tsarnaev brothers.

“The FBI used facial recognition technology to quickly track down and identify the alleged perpetrators,” says electrical and computer engineering Assoc. Prof. Dalila Megherbi. “The Boston Marathon attack is a timely but painful reminder of the importance of developing accurate, reliable and robust facial recognition algorithms.”

Megherbi, who is an expert in digital image processing, computer vision and artificial intelligence, is director of UMass Lowell’s Center for Machine/Human Intelligence Networking and Distributed Systems (CMINDS).

Facial recognition programs work by matching selected features from the subject’s face, such as the eyes, nose, cheekbones and jaw, in a digital image or video frame to records on file, such as driver’s licenses or passport/visa applications. 

Megherbi and computer engineering graduate student Iliana Voynichka have been conducting research at CMINDS, focusing on facial recognition, especially with facial expressions or disguises that vary over time. The two were invited to present their findings at the IEEE International Conference on Technologies for Homeland Security on Nov. 12 in Waltham. Their work, entitled “Analysis of the Effects of Image Transformation, Template Selection and Partial Information on Face Recognition with Time-Varying Expressions for Homeland Security Applications,” won the conference's Best Paper Award.

“The law-enforcement community recognizes that face recognition plays a crucial role in surveillance at airports, federal buildings, border checkpoints and other places as it doesn’t require the subject’s cooperation — that is, the images can be obtained from a distance, without the subject being aware,” Megherbi explains. “The challenge is trying to achieve a balance between protecting a person’s privacy versus public safety and security.”

In the case of the marathon bombers, the FBI was able to assemble a complete picture of the suspects and recreate the timeline of their activities and locations based on the scores of photos submitted to the agency by people as well as from security cameras mounted in restaurants and office buildings around the crime scene.

“This was a lot of work considering that the images show different angles of the suspects’ faces, and the images themselves were of different qualities, resolutions, scales and lighting conditions,” she notes. “Also, both suspects were wearing baseball caps and one had sunglasses.”

Megherbi and Voychnika’s research shows the effects on facial recognition accuracy of some selected  factors, including image facial registration with or without off-the-plane image rotation, the type and number of the individual’s face templates chosen and the type and increasing amounts of partial facial information contained in face images. 

“Hopefully, our findings will help in building better face-recognition systems so we might be able to prevent future tragedies like the Boston Marathon bombing from happening,” says Megherbi.