Improving Interactive Image Segmentation via Appearance Propagation

Abstract

We present Propagate-and-Graphcut (PG), a system that improves interactive image segmentation by leveraging the appearance propagation research. Users efforts are minimized, and inaccurate inputs are handled. We show experimentally that PG is capable of handling complex scenario settings and outperforms previous approaches.We suggest that PG could serve as an alternative to other general interactive image segmentation methods under most circumstances.