摘要

A novel approach based on spatial-temporal information fusion is proposed to segment moving objects in video sequences. In the proposed approach, temporal and spatial information are integrated gradually by a region binding process during the segmentation, which is different to the process of combining temporal and spatial segmentation results. The significance of the proposed region binding algorithm is that regions are represented and characterized distributively. In the first stage of the proposed approach, current frame is segmented into many small regions primitively. Then, these small regions are bound to form the Binding-CCores ( BC). Finally, the rest regions are bound to their neighboring BCs under strong or weak rules to get the final object regions. Experimental result demonstrated the performance of the approach.