A survey on visual adult image recognition

Abstract

We provide an overview of state-of-the-art approaches to visual adult image recognition which is a special case of one-class image classification. We present a representative selection of methods which we coarsely divide into three main groups. First we discuss color-based approaches which rely on the intuitive assumption that adult images usually feature skin-colored regions. Different ways of defining skin colors are described and example classification frameworks built on skin color models are presented. Another main group of approaches to adult image recognition is based on shape information which usually also exploit color information to find skin-colored regions of interest. Color and texture features are often used to augment such shape features. Finally we introduce approaches based on local feature descriptors. © 2012 Springer Science+Business Media, LLC.