The Pierce County Sheriff’s Department is testing computer software that checks surveillance photos from crime scenes against mug shots of potential suspects. So far deputies have made two arrests.
 Pierce County Sheriff’s Department forensic services supervisor Steve Wilkins tests face-recognition software. It matches surveillance photos to mug shots. (AP Photo) |
By Stacey Mulick
The News Tribune
TACOMA, Wash. — The forgery and theft case had victims, a witness and decent surveillance images from an ATM. What it didn’t have were any leads on who committed the crime.
But instead of being tossed aside, as happens in so many property crime cases, the ATM images were e-mailed to Steve Wilkins at the Pierce County Sheriff’s Department.
Wilkins, the department’s forensic services supervisor, picked the clearest image and used new facial recognition software to compare it to 16 years’ worth of prisoner mug shots taken at the Pierce County Jail.
Within 15 minutes he’d found a match.
Detectives followed the new lead and eventually arrested Susan Bennett, who was charged in October with 11 crimes in connection with the ATM thefts. She pleaded guilty Thursday and was sentenced to 91/2 years. Half of that will be served in prison, and half in community custody under Department of Corrections supervision.
The match was the first for the Sheriff’s Department’s six-month pilot project with Sagem Morpho Inc.’s new facial recognition software, MorphoFace.
The computer program runs with Hollywood-like ease and, with a click of a button, compares a suspect’s image to a database of thousands of known Pierce County offenders.
“It’s really cool,” said Wilkins, who wants to buy the final version of the software when it hits the market in January and add it to his forensic tools.
Detectives and Tacoma-Pierce County Crime Stoppers officials are revisiting unsolved bank robberies, ATM scams and other crimes that have good surveillance images to see whether the program can crack the cases.
Suspect photos from a handful of criminal cases have been run through the software but didn’t generate any hits.
Nevertheless, officials are optimistic the new software would solve crimes in the long run.
“We are pretty excited about this,” said sheriff’s detective Sgt. Tim Kobel, who supervises the Property Crimes Unit. “We think it is going to turn some cases for us and we’ll make some arrests.”
The Sheriff’s Department paid Sagem Morpho Inc., a French biometric technology company whose North American headquarters is in Tacoma, $18,000 to test the software from June through December.
The department is one of two law enforcement agencies in the country piloting the MorphoFace software, which is on a computer in the Forensic Services Unit in the County-City Building in Tacoma.
Other major law enforcement agencies in the state, including Tacoma, Seattle and Spokane police departments, don’t use such software.
But facial recognition software is not new on the national and international law enforcement stage.
Tampa, Fla., used the software in 2001 during Super Bowl XXXV. It scanned the crowds as spectators went through the turnstiles and compared the faces with images in a database of known felons and suspected terrorists. The scans spotted 19 people with criminal records, but none was sought by the authorities, according to The New York Times.
After the 9/11 terrorist attacks, several airports looked at installing facial recognition cameras as a security measure. The ACLU opposed the idea, arguing that the technology was intrusive and ineffective. Its use also raised privacy concerns.
The Sheriff’s Department is not using the software to scan crowds and look for felons. It’s intended to be an investigative tool for detectives, said Eric Hess, Sagem Morpho’s North American product manager for facial recognition and iris recognition.
“This is a tool to help them identify likely suspects,” Hess said. “There is no way to lift a fingerprint off an ATM.”
Pierce County is testing the software on two fronts – to gauge its accuracy and to try to solve open criminal cases.
To test the software’s accuracy, Wilkins checks whether it can match current mug shots of repeat offenders currently in the jail with their previous booking photos.
He goes through the daily bookings and selects the men and women who have been locked up before. He takes their most recent mug shots and uploads them into MorphoFace.
The program is asked to find possible matches in a database of more than 479,000 mug shots of people booked into the jail, the Remann Hall juvenile jail and the Puyallup City Jail since 1992.
MorphoFace uses algorithms to measure the location of a person’s eyes and builds a model of the face that is compared to the mug shots in the database.
“It will recognize unique patterns in each person’s face,” Hess said.
The software returns a candidate list for Wilkins to study further. The best comparison is listed first, with lesser matchups following.
“You can see this progression of their ages,” Wilkins said.
The process is similar to how technicians examine fingerprints.
They pinpoint unique characteristics on a suspect’s fingerprint, then run the print through a database to see whether it matches known offenders. Forensic technicians further study any candidates for a match.
Wilkins has found that the facial recognition software spots who it should about 90 percent of the time. Among those matches, 99 percent of the time the first candidate is the right one.
As for solving crimes, property crimes detectives have submitted nine cases for evaluation. Other images have been submitted but were not suitable to run through MorphoFace. The picture quality and resolution were poor or the camera angle was bad. The images have to be of decent quality.
“It has to be reasonable,” Kobel said. “We look at some images, and you can’t tell whether they are male or female, black or white.”
Two have resulted in hits and, later, arrests, Kobel said.
The Bennett case was the first. Investigators also were able to identify a man who presented a false driver’s license with a forged check.
“I feel extremely fortunate to have gotten this one,” Wilkins said of the Bennett case.
The 40-year-old woman is accused of withdrawing cash and buying items using other people’s debit cards. She rang up several thousand dollars’ worth of charges on ATM cards from Franklin Pierce School District booster club members and a Boeing Employee Credit Union member, prosecutors allege.
Several of the ATM transactions were captured on surveillance video. But the photos were not ideal, and investigators had no clue who the thief was.
Kobel sent the images to Wilkins, knowing the department was testing the facial recognition software.
“Otherwise we would have had to return to the whole gumshoe routine,” said Kobel, who sifts through 60 to 90 property crimes cases a day and determines which are assigned to detectives for follow-up. “I don’t have time for that.”
Wilkins picked the best surveillance shot, cropped it and rotated the image. He ran it through MorphoFace. He quickly matched the image to a previous booking mug shot of Bennett, who was convicted last year of identity theft charges.
Wilkins relayed the results to detectives, who continued their investigation and eventually made an arrest.
“If we didn’t have a tool like this, that case wouldn’t have been assigned” to a detective, Kobel said. “This is going to be a huge tool for us.”
How it works
• A suspect’s image is uploaded onto a computer.
• The software uses algorithms to reconstruct the suspect’s “face.”
• The program starts by measuring the distance between a person’s eyes, which are a similar distance apart on most faces. From there, the program builds the rest of the face, recognizing patterns in faces that are unique from one individual to another.
• The program then searches the database for other faces with similar patterns.
• A candidate list is returned. A forensic technician looks at the specifics – the size of the ears, curve of the lips, location of freckles – to determine whether it’s a match.
Copyright 2008 The News Tribune