Oak Ridge National Laboratory (ORNL) staff members from the Image Science and Machine Vision (ISMV) and Applied Software Engineering Research (ASER) groups have been performing volunteer work with the Knoxville Police Department and Internet Crimes Against Children (ICAC)[1] to develop a software tool called Artemis to combat child pornography and child crimes. Artemis automatically examines the contents of a suspect’s computer to identify child pornographic images, text, and related materials based upon hash values, image content, text, and more. It ranks the files based upon likelihood of being child pornography and allows the examiners to designate which files they deem to be child pornography with a single click. This process automatically generates a report on the illegal material, saving valuable time both in the lab and in the field. The ranking system reduces the number of files to be inspected by 10 to 20 times, which, for a computer with half a million images or more, can amount to days of time. Furthermore, Artemis can examine 25 or more files per second on a standard desktop computer, which is faster than humanly possible, allowing for fast onsite results. The current Artemis tool can be run either off a bootable Linux thumb drive or under the Windows OS. Artemis can be run on either a law enforcement or suspect’s computer and is easy to use even for those not trained in computer forensics or child pornography. Artemis is currently used in the field by the Knoxville Police ICAC Group and Blount County Sheriff’s Department, who are participating in the development and testing of the tool prior to further dissemination.
“Artemis decreases the time to perform an examination and create documentation of probable cause for an arrest by more than half, and the onsite examination ability allows for an immediate arrest on site.”
- Tom Evans and Mel Pierce, Knoxville Police Investigators
This proposal seeks to develop the Artemis tool in a few key areas that will significantly enhance its capabilities and usefulness. Extending the existing image content evaluation techniques to include movie files will substantially improve the tool. Combining the current face detection capabilities with face age estimation software from the University of North Carolina will make this the first tool capable of distinguishing between child and adult pornography based upon image content. Training new feature detectors will allow for automatic detection of genitalia and other pertinent image contents in addition to faces, further expanding the ability to identify pornography. Integrating database management tools to synchronize the results of examination with other databases will increase the tool’s ability to identify known child pornography and share information across independent law enforcement departments.
