DecNec
Detecting CSAM Without the Need for CSAM Training Data
Detecting child sexual abuse material (CSAM) poses a major challenge for law enforcement agencies worldwide. While known CSAM, such as material seized during an investigation, can be identified using hash matching (e.g., perception-preserving hashes like PhotoDNA) , the detection of unknown CSAM content requires the use of advanced methods - particularly those based on AI. However, the use of CSAM training data in AI is extremely problematic due to its highly illegal nature (see StGB §184b+c), particularly for research institutions.
This is precisely where the DeCNeC project comes in, focusing on the detection of unknown CSAM without resorting to CSAM data. As part of the project, novel approaches in the field of computer vision are being researched and developed. The core idea is to combine separate concepts to enable CSAM detection. To this end, classifiers are trained separately on adult pornography scenes and on harmless images of children in everyday situations and then fused to identify potential overlaps (and thus CSAM).
In addition, DeCNeC is dedicated to researching image enhancement techniques to enable effective processing of even low-quality images. In parallel, solutions are being developed to address challenges such as problematic depictions (according to the COPINE scale, a system for classifying the severity of depictions of child sexual abuse). In close collaboration with law enforcement agencies, DeCNeC aims to develop ethically sound and practical detection methods that can precisely and effectively support investigative work.
Seeking partners!
To ensure that the DeCNeC project produces AI models that are as practical as possible, collaboration with a State Criminal Police Office (LKA) is essential. The LKA’s role will be to create suitable CSAM datasets so that the AI models developed in the project can be evaluated (by the LKA). In the second phase of the collaboration, the LKA will provide statistics regarding the evaluation of the AI models so that they can be continuously optimized.
Participating State Criminal Police Offices will have free access to the specialized AI models throughout the entire project duration (4 years), which deliver more reliable and explainable results than paid competing solutions.
Interested parties are welcome to contact the project management team, including for more detailed information on the structure of the collaboration.
