Image processing techniques to detect and filter objectionable images based on skin tone and shape recognition

Abstract

In this paper we describe image-processing techniques to detect and categorize potentially objectionable images based solely on the image content. The techniques include a novel method of matching skin tones based on a fuzzy classification scheme and shape recognition techniques to match faces and other elements of the human anatomy. In particular the color matching technique is significantly more accurate than existing methods documented in literature. These techniques offer a means of filtering images, based solely on graphical content, for a new generation of home Internet appliances.