Graph-Cut Based Background Subtraction Using Visual Hull in Multiveiw Images

A graph-cut method has been successfully used in many applications for image segmentation. However, it needs lots of time and user intervention. In case of multi view image (MVI), it is especially hard to segment all images in a short time because of numerous images in MVI. In this paper, we describe a new technique for multi view image segmentation, which needs minimum user intervention and provides fast processing time. The user marks certain pixels as "target object" or "background" to provide a constraint for segmentation to only one of the MVI. The seed information is propagated to all images in the MVI. In this step, we can acquire tentative segment result and then apply them to reconstruct the 3D model which exploits the visual hull. After the 3D model is reconstructed, segment error that is found located out of foreground is eliminated. Although visual hull has a shortcoming that cannot represent whether the object is convex or concave, tentative segment result is easy to use and proven to be enough as our proposed method. We can acquire final segment result in a short time by integrating these two simple methods. According to the experiments, our method shows better performance in terms of processing time and minimizing user intervention.