Skin Segmentation Based on Double-Models

In the process of skin tracking of video, Skin color model often can´t get expect affection in some environment. This paper brings out double-models based on analysis of color space and traditional skin color models. Using the interaction of two models, We can adjust their parameters which can adapt to the environment. The skin color modeling include two steps. First it is stat stylebooks´ parameters. and Second is dynamic parameters´ adjustment. This paper chooses ellipse clustering model and SGM as basic models and YCbCr color space to describe the pixels. We use the minimum outer rectangle to get ellipse clustering model´s static parameters, and use the Fisher threshold rules to compute the SGM´s chromaticity diagram. When adjust the dynamic parameters, we look on the result of these two models recognition based on static parameters as some stylebooks. From the result of experiment, double-models have the advantages of strong robust and can boost the segmentation efficiency.