A New Way to Detect Digital Child Pornography?

The fight against child pornography could be getting a new high-tech tool.

To help law enforcement with the task of analyzing a suspected child pornographer's computer, new software developed by a computer science professor at the Polytechnic Institute of New York University brings deleted photographs back from the computer's trash and searches them for potentially explicit images of children and differentiates them from images of adult.

The program scans for faces of children, nudity and other features to help flag images that could possibly be illegal contraband.


"It uses machine-learning algorithms to distinguish child from not-child," said Nasir Memon, a professor of computer science who created the program with his students.

"Machine learning" refers to a process by which a program learns to identify certain kinds of images by processing other similar images.

"[The program looks] at the face, skin, nudity, other features potentially that combine together, to pull out the most likely images which could be problematic," he said.

The program was designed to help law enforcement, defense and intelligence officials and private investigators hired by the private sector, Memon said.

As law enforcement grapples with increasing volumes of digital child porn, child advocates say technology that can help streamline the identification process is becoming more valuable.

"It is a struggle for law enforcement who are working child porn cases," said Michelle Collins, vice president of the National Center for Missing and Exploited Children's exploited children division. "Over the last few years, the size of the child porn being seized has increased dramatically."

Program Computes Distance Between Eyes and Nose to ID Children

Memon said the program computes the distance between a person's eyes and nose and other facial features to separate children from adults, but low light and non-frontal photos can skew results.

The program is about 70 percent accurate in identifying images of children, Memon said, but he added that even that success rate could be helpful in narrowing the field for investigators building a case against a child porn suspect.

Through Digital Assembly, a Brooklyn, N.Y.-based start-up created by Memon and two of his students, he previously released a version of the software that recovers deleted and fragmented digital images. The newest version of the software, called Adroit, will launch later this month and includes the new filtering technology.

In addition to scanning for potentially explicit images of children, he said the program can search for explicit images of adults, photographs of a particular person and indoor or outdoor photos.

Child Advocates: Size of Porn Being Seized Is Increasing

With the proliferation of more affordable computer storage and more people using broadband, the average collection of child porn seized by law enforcement is growing each year, according to Collins.

Collins said the National Center for Missing and Exploited Children helps law enforcement match found pornographic images with the people who created them to help the children victimized in the process.

Through the Child Victim Identification Program (CVIP), her group helps prosecutors by examining images and videos to prove that a real child is depicted in each pornographic photo. Each known pornographic image is assigned a unique identifier, which is saved in a database.

Using those identifiers, law enforcement can run software to search a suspect's computer for known child porn, but not new child porn, forensics experts said.

Detecting Child Porn Is Especially Challenging

Victor Fay-Wolfe, director of the Rhode Island Digital Forensics Center, said his center will release a downloadable program next week that is similar to Adroit that could help law enforcement scan a hard drive for porn.

With funding from the National Institute of Justice, the program examines images for skin tone, edges that indicate human forms and other features.

"All of those weighted together allow the software to determine together if it's porn," he said.

But he added that automating child pornography detection has proved especially challenging, in part because child porn laws not only apply to offenders, but those developing technology intended to help prosecute them.

"Child pornography is a different story," he said. "We're finding it to be an extremely difficult problem."

Detection software needs a straight on image of a face, which you don't often get in child pornography, he said. Photos of genitalia would be more effective, but using those is illegal.

"That's the best we can write without having contraband," he said.

Digital Forensics Expert: Some Detection Is Better Than None

Fay-Wolfe said programs created to detect child pornography are at most 60 percent accurate. But even that could potentially help law enforcement, he said.

"Even 50 percent accuracy is a big savings of time to them," he said. "Some detection is better than none, when they have nothing to help them."

Still, though technology may help an investigation into a child porn suspect, law enforcement officers say that it's ultimately the human investigators who makes the biggest difference.

"The skill and experience of our investigators, along with the intelligence we develop, is our strongest asset in pursuing these criminals," said Peter Grossgold, an FBI special agent who supervises the squad that investigates child pornography cases in the New York office.