Automatic extraction of moving objects for head-shoulder video sequence

Recently, video object extraction has received great attention because it is a critical technique in object-based video processing. This paper presents a temporal-to-spatial segmentation technique to extract object from a video sequence. The temporal phase employs a simple blockwise temporal-activity measure to approximately locate the object boundary. And then a block-based maximum a posteriori (MAP) scheme, which exploits spatial features of image blocks around the approximated boundary, is adopted to refine the temporal segmentation result. The proposed technique achieves good segmentation quality with very low computational cost for head-and-shoulder sequences with static background.