Skin-color detection algorithm based on body-localization and dynamic skin-color threshold

With the popularization of Internet, research on content-based erotic image filtering becomes a new hot research spot in recent years. In erotic image filtering, skin-color detection is one of the key technologies. Aiming at the problems in current approaches for skin-color detection, an approach for skin-color detection is proposed, which is based on human face and body positional information. Firstly it carries out human face detection using AdaBoost algorithm. Secondly it establishes body model based on human face information and realizes human body approximate localization. Finally, aiming at the deficiency of traditional skin-color detection algorithms, the absolute and relative position information algorithm is proposed, in the algorithm, each picture element has different skin-color threshold value according to human face and human body information. Experiment results show that the proposed approach is feasible and valid. It has a good result especially when human face occupies a large proportion and the background has much skin-like color in the image.