Human segmentation based on GrabCut in real-time video sequences

In this paper, we present a fully-automatic human segmentation method without iteration in video sequences. To segment human body accurately, we adopt coarse-to-fine approach using human detection and background subtraction. HoG-based method is used to detect human ROI. Background subtraction is used to subtract subject image in human ROI and skeleton image. The human ROI, the subject image, and the skeleton image are initialization values of GrabCut. The initialization values provide more accurate foreground and background information to GrabCut. Therefore the proposed method can segment human silhouette accurately enough to apply in video analysis without iteration. Experimental results show that the proposed method can be achieved better performance than GrabCut in real-time video sequences.