Accurate face segmentation based on fusion of improved GrabCut and ASM

In this paper, we propose a novel algorithm for accurate face segmentation in frontal images. GrabCut has been widely applied in segmentation, however, it needs user interaction and suffers from the inaccurate foreground model. Since Active Shape Model (ASM) is able to approximately locate the face region, this paper introduces ASM into GrabCut to avoid user interaction and improve the foreground extraction. It includes three phases. First, landmarks of face images provided by ASM, instead of user interaction, are utilized to locate the face region. Second, based on the facial masks from ASM, an improved GrabCut is proposed to conduct the face segmentation. Finally, a bank of Gabor filters of different orientations is utilized to refine the extracted face region. Experimental results on CMUPIE database show that the proposed method significantly improves the frontal face segmentation.