A human skin detector combining mean shift analysis and watershed algorithm

Detecting human skin in color images is an interesting problem due to the potential number of target applications ranging from face detection to the detection of pornographic images. Many approaches have been proposed in literature including color thresholding, statistical skin color modeling and segmentation-based techniques. However, the performances often decrease in the presence of illumination artifacts resulting in sub-optimal results. In this paper, we propose a new method that is adaptive to the image content especially in illumination conditions. For this purpose, we first detect skin seeds that serve as the basis to locate the skin colors with highest density. This step allows to take into account illumination conditions for the last step, namely the growing of skin seeds.