Abstract: Matting technique is a popular method to extract objects from an image precisely. It takes
important advantages in image editing. Most of the existing matting techniques rely on human
interaction. In this paper, a novel automatic matting method is proposed and it is based on moving
saliency objects. Firstly, the input image is segmented into several regions, its saliency values are
calculated, a saliency gravitation model between regions is established, and the moving rules of saliency
regions are constructed to obtain an accurate saliency map. Secondly, the saliency map is obtained and
morphological operators are employed to correct the object regions. The corresponding trimap is
generated automatically. Finally, the object regions are extracted by using learning-based matting
algorithm. The experimental results show that the proposed method is more accurate and effective
without human interaction or other auxiliary information.

Key words: image saliency, moving saliency regions, automatic matting