A fast boundary-preserving method for image object segmentation

A fast boundary-preserving method for interactive image object segmentation is proposed. It is based on the graph-cut framework, and establishes a two-step optimization procedure from global segmentation to local refinement. In the global stage, pixels are replaced by superpixels which preserve necessary image boundary properties and largely reduce the computational complexity. The shape prior knowledge of object is introduced and coupled with traditional appearance and gradient information to guarantee the region and boundary wholeness of global result. For a boundary segment which needs to be refined, the local refinement uses the local appearance model updated by local sampling to rectify errors within its local area experimental. Results prove the correctness and effectiveness of the proposed method.