Face and silhouette based age estimation for child detection system

Abstract

The problem of age estimation based on facial images is a well-known computer vision task that is widely applied in identification systems. With help of the special Dyzurnet.pl unit detecting Internet content, related to sexual abuse of children we slightly redefined a problem. Our Convolutional Neural Network (CNN) solution is focused on infants and prepubescents recognition and in the particular age ranges can be considered as the-state-of-the-art in children detection.Silhouette-based age estimation is often concentrated on the human gait or body proportions analysis. Single image age estimations on the dressed (fully or partly) body are not typically researched because of a lack of properly labeled data. In our work, we present the method used to train image preparation and the final effectiveness of age estimation of that kind.The proposed solution is a part of the system for responding to threats to children's safety in cyberspace with particular emphasis on child pornography