Griffeye has teamed up with the New York University Tandon School of Engineering to start building novel tools to help identify and rescue victims in child sexual exploitation material. Sponsored by the US National Institute of Justice, development will take three years with tools being released to the user community over that time. Griffeye Director Johann Hofmann tells us more.

What’s the background to the project? Why did you join forces with the NYU School of Engineering?

As we know, investigators working to identify victims in child sexual exploitation material must deal with an overwhelming volume of image and video files for each case. We’re talking potentially millions of images and many thousands of hours of video per case. The team at NYU Tandon, led by Nasir Memon Professor of Computer Science and Engineering, has developed sophisticated techniques to aid investigators in prioritizing data by combining various concepts of classifying image and video content in great detail. These concepts could include the presence of skin tones and map regions of body parts to determine if a subject is nude. When combined with facial recognition to detect children’s faces, this can identify sexual exploitation material with high accuracy. By integrating these advanced machine learning techniques into our Analyze platform, we’re looking to automate and greatly speed up the analysis. Which, in turn, leads to victims being identified and rescued much quicker.

Speed is clearly of the essence in these cases, isn’t it?

Yes. With the scale of the problem, it’s all about speed and accuracy if investigators are going to get results and rescue victims. Firstly, by leveraging robust techniques already available in the Analyze platform to automatically identify and filter out non-pertinent data, we can save the investigator manual labor by reducing caseloads with up to a certain level (we are sometimes hearing 90%). These new deep learning techniques developed by NYU can automatically aid this process even more. Only leaving the investigator with a small subset of data left to investigate. Then we can prioritize the remaining files with the most probable value to the case. From there, using our open API to bring in whatever tools they need, investigators will be able to detect critical clues in individual files. And last but not least the platform can also help them identify relationships between different entities so they can build and crack the case.

READ MORE: Co-development with New York University

How have law enforcement agencies responded to this?

It’s big news for them. Not only has the US National Institute of Justice given the project serious financial backing but we’ve also got the involvement of the US Department of Homeland Security Child Exploitation Investigations Unit during the development process. The latter will make sure that what we develop also works in practice and that it is used by as many as possible. It’s really great to see what we can achieve when we bring the best minds and tools together to solve a huge and complex problem.

“Working together to get results” seems to be a recurring theme at Griffeye. Is that your philosophy?

You can call it a philosophy but I think it’s just we know that is what is needed. We know we need an open platform to make sure users get the best tools for the job. And we know that by using Analyze as a platform for development with expert partners we can make even greater strides.

Talking of the users, when can they start using the tools that are being developed?

The whole development project is scheduled over three years. But, of course, as soon as a tool is developed we’ll release it to the user community. We want to help achieve results out in the field as soon as possible.

PRESS RELEASE: New forensics tools will speed the identification and rescue of children pictured in child sexual exploitation material