A Comparative Study of Tools for Explicit Content Detection in Images

Adrien Dubettier (1) , Tanguy Gernot (1) , Emmanuel Giguet (1) , Christophe Rosenberger (1)

1 Equipe SAFE - Laboratoire GREYC - UMR6072

Cyberworlds offer a vast quantity of knowledge and services on all topics for Internet users. The protection of children is an important issue on Internet and could be solved by detecting automatically explicit content. Another application is to facilitate digital forensic experts when analyzing media such as hard drives to detect child pornography content in criminal affairs. In this work, we focus on images and we study the efficiency of existing methods from the literature mainly based on machine learning and deep learning approaches. We apply a rigorous protocol with significant datasets in order to draw conclusions on the performance we can expect in real conditions. This study shows that this task is not really solved by existing tools. Moreover, the frontier of explicit content is also not always easy to define.

Mots clés: Explicit Content Detection; Digital investigation; Deep Learning; protection of children in Cyberworlds

Domaines: Traitement des images [eess.IV]; Apprentissage [cs.LG]

Adrien Dubettier, Tanguy Gernot, Emmanuel Giguet, Christophe Rosenberger. A Comparative Study of Tools for Explicit Content Detection in Images. 2023 International Conference on Cyberworlds (CW 2023), Oct 2023, Sousse, Tunisia. pp.464-471, ⟨10.1109/CW58918.2023.00077⟩. ⟨hal-04179978⟩