Interactive Foreground/Background Segmentation Based on Graph Cut

Abstract

This paper addresses the problem of natural image segmentation. We develop the interactive image segmentation system and construct two-scale graphs, including region-based graph and pixel-level graph. Building region-level graph is to partition the image into several constituent components. Graph cut is used to extract the foreground object from the image. New data cost functions are defined in the graph cut framework. Under some circumstances, boundary edit based on a pixel-level graph is performed in order to get accurate foreground edges. Experimental results prove that new cost functions are valid and satisfying segmentation results can be obtained by limited user efforts.