A novel matting system using human selective attention

Image matting is an important task in image and video editing. In this paper we propose a novel matting system, which can provide good results with less human intervention. We split the task into two steps: attention-based scribble extraction followed by alpha matting. We use the attention shift trace which is refined in the HSI color space as the useful constraints for matting algorithm instead of user-specified constraints. Distinct from previous approaches, the system presents the advantage of being natural and intuitive to humans and less human intervention.