Automatic video object segmentation based on spatio-temporal information

A novel video moving object segmentation algorithm based on spatio-temporal information is proposed in this paper. In temporal information extracting, the value of background noise variance is estimated by histogram fitting to overcome the shortcoming of setting the value by experience, then the significance test is applied to threshold the difference image and extract the moving areas, furthermore, a symmetrical difference method is adopted to achieve accurate moving object mask. In spatial image information extracting, an improved multi-scale watershed algorithm based on viscous morphological gradient correction and edge value merging is employed to get spatial regions which can solve over segment problems greatly. Finally, video object are extracted by performing double threshold ratio operation on spatial and temporal results. Experimental results validate the proposed algorithm.