A child recognition system based on image selection patterns

In recent years, a market of games and learning websites and apps for children has grown everywhere in the world. With the increasing development of online services on various devices such as PCs, smartphones and tablets, it is important to recognize kids across the web. This is not only required to provide them with more effective services e.g. for education purposes, but also to prevent them from having access to certain content e.g. violent or sexual films. This paper presents our ongoing effort in developing a child recognition system based on the patterns users produce when clicking a set of pictures. We propose a system where users are presented with a set of images chosen from two different pre-prepared categories: children and non-children (adults). The users can click on any of these pictures without any specific rules for this selection process. We implement a web-based prototype of our system and test it with 120 different users (60 adults, and 60 children). The initial results show that children have different image selection patterns compared to adults which enable us to tell them apart from each other.