Skin Detection on Images with Color Deviation

Skin detection is an important pre-process step in Web image filter. Since color modal is based on pixel-level color information, the performance of skin detection on image with color deviation is poor. Automatic white balance (AWB) is necessary to increase skin detection accuracy. Considering that almost all nude images contain human faces, human faces can be used as reference object to perform AWB. To begin with, this paper makes use of Adaboost face detection algorithm to determine face rectangle, then apply a skin detection method immune to illumination to detect skin pixels within the face rectangle. The skin color of face is then used as reference color to perform AWB before skin detection. Experimental results show that the proposed method can increase skin detection precision significantly.