Skin Detection Using a Statistical Color Spaces Fusion Model

Skin detection is the preliminary stage of many computer vision applications. In this paper, a statistical fusion model is proposed for detecting skin regions in arbitrary images. We used conditional random field (CRF) to statistically combine the information of different color spaces and model the spatial relationship between image pixels. The conditional probability distribution of labels (skin or non-skin) given different representations of raw images in various color spaces is modeled. For a test image, the label image is inferred using the learned CRF model. The experimental results verify the success of our proposed method.