Child pornography is a really nasty business. Sexualized images of nude children making their rounds on the internet or stored secretly on private business servers are part of the problem. They make the rounds between people interested in such things, and help keep interest peaked. The images themselves aren't the problem of course. It is what is done to the kids – usually against their will – to produce the images that is the real problem. Those who partake in the finished product often doing so knowing what is involved in making them.

So there is definite interest in tracking such individuals, and tracing the images back to their source. Unfortunately doing so amongst the billions of images uploaded to the net daily is a huge problem. That's before even touching on how to search the millions of files on a school or hospital's private network to find evidence of similar pictures stored there by an employee.

A joint effort by two German organisations hope to have created a new intelligent system to help in the fight. The State Office of Criminal Investigation and the Fraunhofer Institute for Production Systems and Design Technology, both based in Berlin, have collaborated on the development of an artificial intelligence, self-teaching and machine vision piece of software designed to identify pornographic images of children, and only pornographic images of children, amongst the billions of images on the web, or on a private network.

The team headed by Dr. Bertram Nickolay call the new program “desCRY,“ a seldom-heard word in English meaning “to catch sight of, to discern. It is completely automated, rummaging through images on the server it is targeting, or the net in general at lightning speed. Capable of searching still images of all known formats, and video streams as well, desCRY records nothing it finds, letting the images wash over it's 'eyes' numbly, <i>unless</i> it sees something interesting.

In essence, it is programmed to think like a person who is aroused by images of child pornography, for when it finds something that seems to be proportioned like a child, looks like a child facially, and is in an unseemly state of undress and/or a provocative pose, it gets 'excited' and tags both the image and it's location for further investigation.

The heart of the software consists of intelligent pattern-recognition algorithms that automatically analyze and classify images and video sequences. “Technologies such as facial and skin-tone recognition are combined with contextual and scene analyses to identify suspicious content,“ project manager Raul Vicente-Garcia explains. The algorithms use up to several thousand characteristics that describe properties such as colour, texture and contours in order to analyse whether an image depicts child abuse. If the system is run on a standard PC, it classifies up to ten images per second, drastically accelerating detectives‘ investigations.

The software searches all of the files in a computer, e-mail attachments and archives included. Another benefit: Among other features, files can be filtered according to size and type; this makes the work of analysis less time-consuming. What’s more: desCRY offers a wide variety of search options. It can perform content-based data sorting and filtering, for instance. This way, investigators can sort files by person, object or location, for example.

The search result is displayed in an image viewer that can accommodate several hundred photos as tiny icons visible at a glance. Suspicious photos are singled out, for instance by displaying them at the top of the list of results. Investigators can enlarge the images with a click of the mouse; a second mouse click stores them as evidence. Still, the system does not make detectives redundant. At the end of the analysis process, an investigator must determine whether the photos classified as illegal really contain prohibited content.

Detectives of the criminal investigation department are currently reviewing the suitability of desCRY in realistic field tests. The system could become part of the Berlin investigators‘ everyday toolkit by October. And they are not the only ones expected to benefit. The software group SAP has agreed to serve as integration and marketing partner. The group from Walldorf, Germany, has already joined the Fraunhofer IPK in initial discussions with other pilot users internationally.

Criminal authorities in several countries have already shown an interest in the system. As Nickolay points out: “Experts in criminal investigations find our methodology for the identification of child pornography a particularly innovative and promising approach in an international comparison.“

References

Software to prevent abuse at the click of a mouse