Adult image content filtering: A statistical method based on Multi-Color Skin Modeling

Abstract

Automatic skin detection is a key enabler of various imaging applications, such as face detection, human tracking, and adult content filtering. In 1996, the first paper on identifying nude pictures was published. Since then, different researchers argue different color models to be the best choice for skin detection. But, to the best our knowledge, no significant work has been reported previously that attempted to use more than one color model and evaluate the performance for recognizing adult contents. In this paper, a simple statistical framework for recognizing adult images based on an MCSM (Multi-Color Skin Model) is described. From a high level, our approach works in two steps. First, skin regions in an input image are detected using the MCSM. Then these suspected regions are fed into a specialized geometrical analyzer that attempts to assemble a human figure using simple geometric shapes derived from human body structure. Quantitative evaluation shows that our method compares favorably with the state-of-the-art methods in terms of detection rate and false alarm, while reducing the computational complexity by a factor of 1/6 with respect to the Forsyth's method.