Efficient face segmentation based on face attention model and seeded region merging

This paper presents an efficient face segmentation approach based on face attention model and seeded region merging. A face attention model that jointly exploits the information of skin color and eyepsilas position is first constructed to obtain a facial saliency map, which indicates the position of possible faces and is used to determine seed regions. Then a seeded region merging algorithm based on regional facial saliency is proposed to generate a sequence of regions, and the region with the highest regional facial saliency is selected to represent each face. Experimental results on a variety of images demonstrate the good segmentation performance of the proposed face segmentation algorithm.