Análisis de descriptores locales para la detección de menores en imágenes

Abstract

Child abuse material (CAM) refers to photos or videos containing the sexual abuse of a child. This kind of material involves a criminal activity, whose detection may be assisted by automatic tools. Considering a two stages approach, that determines in its first level whether the content is pornographic or not. If the first situation occurs, the second stage is triggered to confirm the presence of any non adult. This work relies on the second stage. Our approach classifies using facial information, an individual as adult or not. For that purpose, different local descriptors are analyzed, in combination with areas of interest, and fusion techniques. The experimental evaluation is carried out on a dataset designed for this particular problem containing a balanced collection of male samples within the range from 12 till 21 years old.