Transferable Belief Model for hair mask segmentation

In this paper, we present a study of transferable belief model for automatic hair segmentation process. Firstly, we recall the transferable Belief Model. Secondly, we defined for the parameters which characterize hair (Frequency and Color) a Basic Belief assignment which represents the belief that a pixel was or not a hair pixel. Then we introduce a discounting function based on the distance to the face to increase the reliability of our sensors. At the end of this process, we segment the hair with a matting process. We compare the process with the logical fusion. Results are evaluated using semi-manual segmentation references.