A Mumford-Shah level-set approach for skin segmentation using a new color space

In this paper, skin region segmentation is performed using a new color space and a level set method based on Mumford-Shah model. The fundamental idea is to divide the whole problem into two main steps according to the hierarchy of image information. We first perform a rough skin classification pixel-wise using a color coordinate deduced by principal component analysis (PCA) technique. The next step is to implementing image segmentation using a level set with Mumford-Shah model. All the regions are then traversed to calculate ratio of skin pixels to the total pixels within each individual region. Those regions whose ratio reaches the statistical amount will then be regarded as skin regions. Experimental results show that the proposed method is robust, and it could reduce false detection caused by noise or illumination variation.