Skin-Color-Based Image Segmentation and Its Application in Face Detection

In this paper, we propose a technique to efficiently detect human skin in color images with complex background and varying illumination conditions. The strategy consists of modeling the distribution of skin colors using a Gaussian mixture model (GMM) and segmenting the image into skin and non-skin parts, using threshold values that are computed automatically by an adaptive technique. Morphological operators are finally applied to refine the final skin regions and obtain candidate face regions. We show that a mixture model of Gaussians can provide a robust representation of the human skin color to accommodate with large color variations. Using images taken from the Internet and in-house pictures containing persons with various skin-color types in different illumination conditions, experimental results show that the proposed method can cope with a wide range of illumination situations and complex backgrounds.