Have you ever used the magic wand tool in Photoshop? If so, it accomplishes 90 percent of the work described in this paper. The only difference is that this work uses finger touch instead of a mouse click, thanks to the prevailing touchscreen technology. I’m not trying to trivialize the work involved; on the contrary, it should remind us of the complexity and importance of performing high-quality image segmentation.

The authors tried a new combination of energy terms in a level-set framework, which was widely explored in the literature after its invention. As stated in the paper, the core contribution is a new model that fuses edge, region, and geometric cues. Among them, I think the geometry energy stands out as the most innovative. It assumes that the user would always pick the center of the object, which plays as the high-level semantics in the segmentation framework. As more and more wearable computing devices become available, I can foresee that human-computer interaction would become so easy that this high-level semantics will be even easier to capture in the future. Imagine: if a device can track your eye movement to determine what should be the foreground of an image, the segmentation task will be less ill posed for the computer. Actually, the word choices, such as “GrabCut” and “TouchCut,” show the trend of transferring some modeling tasks from computer scientists to general human brains.

I think this paper adopted the level-set framework mainly because of the need for video segmentation, as level-set is by nature a good tool for image tracking. If we consider only the geometric term as unique, other models, especially active contours, could be considered for image segmentation by snapping in the same energy functional if appropriate.