As part of the “Bio4ensics” research project, biometrics specialists at Darmstadt University of Applied Sciences (h_da) are supporting the Hessian State Criminal Police Office (HLKA) in the fight against child sexual abuse. Here, they are using new biometric methods for cyber forensics. The challenge lies in the fact that the biometric facial recognition of children is not yet sufficiently advanced and the error rate correspondingly high. This is where the h_da researchers have now achieved a breakthrough: they fed the algorithm with synthetically generated biometric data of children’s faces. Thanks to them, it is now easier than in the past to recognise real children’s faces in the huge flood of data.

By Simon Colin, 2.12.2024

It’s a demanding job: specialists at the Hessian State Criminal Police Office (HLKA) must sift through child sexual abuse material (CSAM) each and every day. Their enduring goal is to rescue children from sexual violence. Important for their investigations, among other things, is the ability to assess how often children suffer from sexual abuse. To do this, the children concerned must be recognisable. In view of the vastly increasing flood of data, the investigators are dependent on being able to recognise the faces of affected children automatically and reliably in the mass of data. Until now, however, this has been problematic: the biometric facial recognition of children is not yet sufficiently advanced and consequently inaccurate and prone to error.

The funding directive “Cybersecurity Research in Hesse” of the Hessian Ministry of the Interior brought the h_da forensics experts on board. As part of the “Bio4ensics” research project, the biometrics team led by Professor Christian Rathgeb from h_da’s Faculty of Computer Science, together with the Hessian State Criminal Police Office (HLKA) and under the strategic leadership of the Hessian Ministry of the Interior, had set itself the goal of improving the facial recognition of children. “There is a real shortage of data in this area,” says Rathgeb, a specialist in digital forensics and a member of the National Research Centre for Applied Cybersecurity (ATHENE). “We are working here with deep neural networks. These are learning systems that need an enormous amount of training data to recognise faces. However, databases with the faces of younger people previously lacked, meaning that recognition performance diminished the younger the person was.”

Although child sexual abuse material contains a lot of children’s faces for training purposes, Rathgeb continues, it is kept under strict control and no one – apart from police investigators – is allowed to view it. This meant that the h_da researchers, too, ruled out working with the material, which is why they decided to create a huge number of children’s faces artificially and use this database to train the algorithm. To make the faces appear more realistic, the h_da researchers generated several, deceptively real facial expressions for each synthetic identity. They started with artificially generated faces of adults and gradually made them younger. And they were successful: the trained algorithm now recognises real children’s faces better than before. “This shows that synthetic photos can improve biometric facial recognition systems and are also a privacy-friendly option in cyber forensics,” concludes Professor Christian Rathgeb.

The biometrics researchers from h_da also studied two other aspects. In child sexual abuse material, it is usually only the perpetrators’ hands and tattoos that are visible. Nevertheless, it is still possible to draw conclusions from this. That is why the h_da team also synthetically produced a vast number of adult hands with various poses and gestures in addition to the children’s faces generated artificially. “In this way, we have managed to develop one of the first algorithms that can evaluate the quality of real photos,” says Professor Rathgeb. The data basis in terms of the recognition of tattoos was also improved.

In addition, the work conducted by the h_da researchers might help to trace children who have been missing for a long time. To obtain a more realistic impression of what the person might look like today, advanced facial recognition could be used to age the child’s face. The latest cyber forensics methods could also help to solve cold cases. The message to the perpetrators is clear: Never think yourself safe, the investigators will never give up.