Skin Color Analysis and Segmentation in Complex Outdoor Background

This paper provides a way of skin detection in outdoor image based on multiple color space. The clustering is good in the color space YCgCr. Firstly, skin colors are projected in the color space CgCr and the fitting of distribution is carried through in order to wipe off a part of non skin color and gain the intersected image as the result of the first detection. Experimental results indicate that this fitting of distribution can have a good effect on reducing a mass of processing pixels of non skin color. Secondly, skin colors extracted in the first detection are projected in the color space GB in order to further wipe off part of the remaining non skin colors that are not reduced in the first detection by the fitting of distribution. Lastly, the relationships among the three components of every pixel of skin colors and non skin colors in the color space HSL are observed and the percentage of pixels corresponding to a certain relationship is calculated, so part of the non skin color can be further reduced based on the differences we find according to the observed relationship. Experimental results indicate that this algorithm has a good recognition effect and small amount of computation, it can be used in skin color detection in simple environment.