Multimodal Periodicity Analysis for Illicit Content Detection in Videos

In this paper, we propose a multimodal approach to illicit content detection in videos. Distribution of pornographic material over computer networks has been taking place since the inception of the internet. Until recently, most of the research focuses on illicit content detection in images and text. Typically it involves robust skin detection, texture characterisation, shape modelling and keyword filtering. Video however provides the opportunity of exploiting supplementary features including audio and motion for additional confidence in the classification process. This work investigates the use of visual motion information and periodicity in the audio stream for illicit content detection in videos.