Investigating child porn is not an easy job. One way to catch the perpetrators is to look for objects that repeatedly appear in photos and videos, but this requires the police to look at thousands of disturbing images. And even then, all you can hope for is that someone recognises a certain watch or doorpost from other images.

‘And they have to do it all manually’, says Fernández. ‘There are two policemen with the police who know all these cases by heart, but obviously they can’t remember everything.’

Abuse

This is why the police turned to the Spanish institute for cyber security, INCIBE, for help. The institute coordinates a European research project. The police asked them if they would be able to develop a method to improve the process.

Fernández started working on research into image recognition software, and on Friday she is getting her doctoral degree for the work from both the University of Groningen and Leon. For her research, she needed to look at pornographic material featuring children herself. Not an easy thing, she concedes. ‘The images and videos I had to look at were really unpleasant. But I had to view them to know what type of image we were dealing with. The quality might be really bad when a web cam was used, for example.’

Fake dataset

Luckily, the real images helped her create a fake dataset. She was able to use that data for the rest of the research.

Her programme enables the police to select a small region of an image. Then, the programme gets to work on comparing the texture, colour, or shape features of that section to the ones of other images. ‘This gives the police a list of photos that match to a certain degree’, Fernández explains. The officers have to personally check whether the match is actually correct.

The first tests of the system were positive, and INCIBE is now working on uploading the thousands of images into the database. Fernández will also be working on improving and expanding the current application.