关键词: 数字化抠图, 多线索, 泊松抠图, 三分元素图

Abstract: In the natural image matting method, there is a problem that the prior knowledge is excessively dependent and the interactive input is complicated. In order to extend the use of natural image matting methods and improve the automation of natural image matting methods, a digital image matting method based on multi-clue information is proposed. First of all, it uses the depth information and visual saliency information of the original natural image to segment the interest region roughly. Then, the morphological expansion and erosion algorithms are used to expand and erase the rough segmentation result, and it can get the  trisectional element map of the matting process. Finally, the fine foreground object matting result is obtained by using the color texture image and the trisectional element map together with the similarity transfer matting method. The experimental results show that this method can not only achieve the ideal matting effect, but also greatly enhance the automation of the natural image matting method.

Key words: digitization matting, multi-clue, Poisson matting, trisectional element map