一种时空信息联合的运动对象分割算法

Automatic Video Object Segmentation Based on Spatio-temporal Information

提出了一种在通用视频序列中联合时空信息分割运动对象的算法。首先,提出匹配加权的全局运动估计补偿算法,以消除动态场景中背景运动对运动对象分割的影响。其次,时域信息提取中,使用基于直方图拟合的显著性检测及对称差分法获得运动对象模板,以克服依据经验设定阈值的缺点并且提高运动对象模板的准确性;空域信息提取中,提出基于粘性形态学梯度修正和相邻区域边缘强度合并的改进分水岭分割算法,以较好地解决分水岭算法的过分割问题,获得有效空间区域分割。最后,利用双阈值比重算法将时域和空域信息结合,提取出运动对象。实验表明,该算法分割结果准确,有效地解决了背景运动、时域信息不准确、空域过分割以及时空信息难以有效结合的问题。

A novel video moving object segmentation algorithm based on spatio-temporal information was proposed in this paper.The algorithm can extract the moving object from the video sequence with static or global motion background automatically.Firstly,an efficient and accurate global motion compensation method was used to change the motion background to static background.In temporal motion information extracting,the value of background noise variance was estimated by histogram fitting to overcome the shortcoming of setting the value by experience,then the significance test and the symmetrical difference method were applied 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 was employed to segment moving regions which can solve over segment problems greatly.Finally,video object was extracted by performing double threshold ratio operation on spatial and temporal results.Experimental results validate the proposed algorithm.