Winnipeg police combat online child porn with “image matching” software

Police anti-child abuse task force investigators often have to sift through millions of seized photos and videos in their effort to crack child porn rings.

The process, when done manually, is hugely time consuming and “devastating to one’s soul,” as one veteran anti-child abuse expert puts it.

But now the Winnipeg Police Service’s Internet Child Exploitation (ICE) unit has found a way to automate the process, saving investigators a colossal amount of time – and heartache – in the bargain.

The force is the first Canadian police department to use “image matching” software to fight online trafficking in child porn.

Dubbed LACE (or Law Enforcement for Child Exploitation) the software tool drastically cuts down the time investigators need to spend analyzing seized images. It does this by automatically weeding out photos and videos already present in the station’s database.

LACE has been developed by BlueBear Law Enforcement Services Inc., a software company in Gatineau, Que.

As previously logged images account for 74 to 90 per cent of confiscated abuse material, the tool can shave off investigation time by three quarters, and shorten the time officers spend watching horrendous images from weeks or months to hours and days.

The program – initially deployed on trial basis by the ICE unit in July of 2007 – first scans images for descriptive elements, such as pixel counts, colour, brightness, size and other patterns. It then analyzes the images based on these parameters, categories them and determines matches to eliminate duplicates.

“This means if a photo is already in our database, I won’t need to look at it again even if it’s part of recently seized batch that I’m investigating,” said Det. Sgt. Rich Lamire of the ICE Unit.

Prior to the deployment of LACE, officers manually viewed and logged each seized image.

Lamire, 15-year veteran of the force, who has been with the unit for about two-and-a-half years now, says apart from the time saving advantage, LACE reduces the emotional toll on the investigators who have to review such images.

“An important part of our responsibility is identifying the victim and suspect. That means going through anywhere from 300 to a million juvenile sexual abuse images.” This task, he said, can be very emotionally unsettling, especially for officers who have kids of their own.”

When images are automatically sorted by the program, investigators no longer need to re-classify previously logged items. This way they can devote more time to conducting other vital investigative tasks.

Potentially, Lamire said, LACE will also enable faster compilation and presentation of evidence for disclosure purposes and court cases.

Witling down the volume of seized images has become extremely vital to investigative work over the years, according to a former Toronto Police detective who specialized in child abuse cases.

When images showing the sexual abuse of children began flooding the Web in the late 1990s and 2000, investigators working on a case often viewed 50 to 100 photos, said Paul Gillespie, now head of Gillespie Consulting, a Toronto-based an anti-Internet child abuse training and consulting firm.

Gillespie helped kick-start the development of Microsoft’s Child Exploitation Tracking System (CETS). It’s an analytics tool that gathers Internet data, and isolates patterns that help law enforcers track down child abusers.

Sometime between 2002 and 2003, that number of images investigators would found jumped to around 20,000, and now it’s typical of authorities to seize images in excess of a million, Gillespie said.
“The reality is that up to 90 per cent of seized images are photos have been circulating on the Web for years, or have been already logged in a police database somewhere,” Gillespie said.

“If you can prevent an officer from viewing these photos again, you save can enormous amount of time and spare that officer an experience that can be devastating.”

The emotional impact sometimes remains unnoticed for years and may not even be known to the officer, but in some cases investigators have had to seek therapy, the former detective said.

There are numerous image-matching programs on the market, according to Gillespie. Most of these tools use what’s called a “hash function” to turn data into a relatively smaller identifying number that serves as an indexing identifier – also called a hash value.

But the problem with tools that solely rely on hash values is even minor changes to pixel content or the assigned tag name on an image renders that image unrecognizable.

“The program is likely to view that image as a new one even if it’s a slightly altered version.”

LACE, on the other hand, relies on a “soft match” method, which scans for multiple factors to determine whether or not two images are alike,” says Antoine Normand, CEO of BlueBear.

“Hash value-based programs accurately match only five to 10 per cent of scanned images,” he said. “LACE is able to match in excess of 80 per cent.”

The program looks at factors such as pixel changes, colour, brightness, contrast and other features of the image. He said LACE’s algorithms can also be altered for biometric extraction purposes. Such facial recognition systems are being used for national security purposes by some countries.

“With a few adjustments, we can enable the program to look for biometric signatures which can be useful recognizing faces and identifying victims or suspects,” Normand said.

Share on LinkedIn Share with Google+