Interactive Segmentation as Supervised Classification with Superpixels

Image segmentation is a task of extracting desired objects from an image. Though automated image segmentation have been widely researched for decades, the results of automated segmentation is not yet satisfactory enough. One reason is that the “desired objects” in image segmentation are quite subjective so that there could be different objects desired for the same image depending on the purpose of task. As an alternative, an interactive segmentation takes ths human inputs called “seeds” to capture the desired objects by specifying the object and background from them. As a semi-automatic segmentation technique, interactive segmentation has also been actively researched in certain applications including medical image segmentation, which especially requires human expert’s knowledge and interaction to complete the task.