An Iterative Bayesian Approach for Digital Matting
Hang Chang, Qing Xin Yang, Chunhong Pan · 2006
This paper proposes a new iterative approach for digital image matting. It combines pre-segmentation and matting into an unified approach and extracts good matte iteratively within a well-defined Bayesian framework based on a few user strokes on foreground and background regions. This method does not need a well specified trimap, which refers to a pre-segmented image with definitely foreground, definitely background and unknown regions, and can therefore efficiently handle the images, in which the trimap is very hard to create even manually. Experimental results show that, compared with previous approaches, our method is more convenient and robust especially for images with large amount of holes or with foreground objects containing large portion of semi-transparent parts.