In computer vision, you model pixels as graph vertices. Edge weights represent similarity between pixels (based on color, texture, etc.). The goal: split pixels into foreground and background. You add a source (foreground) and sink (background), connect them to pixels with weights based on user hints or color models.
Then compute minimum - cut. The cut partitions pixels into two regions. Pixels in are foreground, pixels in are background. The cut follows natural boundaries where pixel similarity is low.