New software matches sketches to mug shots

Software now picks mug shot from sketch // Source: wareground.com

Law enforcement officers often rely on victims and eyewitnesses to help them identify possible suspects. When the crime is serious, such as the Oklahoma City bombing in 1995, forensic artists work with those witnesses to draw the details that can be used to speed up the capture of a perpetrator. Now, researchers at Michigan State University have produced a software program that “reads” facial details and matches them to mug shots in a central database.

Security Products reports that MSU Distinguished Professor of Computer Science and Engineering Anil Jain and doctoral student Brendan Klare have developed Local Feature-based Discriminant Analysis (LFDA).

In MSU’s Pattern Recognition and Image Processing Lab, Klare collected 159 forensic sketches from the FBI and Michigan State Police. The study also required that the suspects depicted in the sketches were apprehended and their mug shots available from a law enforcement database. Using techniques that describe lines and shading in small patches of sketches, Klare was able to develop the algorithms for the software in about one year. He used a software development kit from Cognitec, which develops face recognition technology, to build LFDA. “We use their system for eye detection; it works with sketches, too,” Klare explained.

Jain, director and founder of the PRIP lab, said that the software finds high-level features, such as the structural distribution and the shape of the eyes, nose and chin, from both the sketch and the photo.

LFDA was able to identify the correct person 45 percent of the time, Klare said, noting that one of the limitations of the study was the amount of available data. “We need a forensic sketch and the picture of the person captured. These are hard to find.”

Sheila E. Meese, forensic scientist/forensic artist for the Michigan State Police Lansing Crime Lab, provided some of her composite drawings and related mug shots to the project. Michigan State Police currently uses S.N.A.P., or the Statewide Network of Agency Photos, which contains the digital image database of mug shots, scars, marks, and tattoos. Not forensic sketches, however.

Meese said LFDA would have great value for law enforcement. “Anything that can help match my composite to a suspect would be a benefit for everyone involved. It would help bring comfort to the victim, help close an open investigation, increase the success rate of the forensic artists, help the community by getting another bad guy off the street, etc.,” she explained.

In addition, such a software program incorporated statewide would provide significant time savings. “For instance, a criminal is arrested in Detroit, gets out of prison and moves to Flint where that police force does not know him. While in Flint, he commits another crime where a composite drawing was done of him. That composite drawing can now be entered into this program and this criminal could be identified much quicker and easier by matching the previous mug shot to the composite,” Meese said.

While Klare said the researchers’ “ultimate goal is to get this in the hands of law enforcement,” much work remains to be done. He has applied for a two-year grant from the National Institute of Justice to further refine the accuracy of the algorithms, develop a prototype system and conduct field testing in Florida.