中文摘要：

针对目前许多视频对象分割方法中分割边界不精确、遮挡和不规则运动问题解决效果不好等问题，提出一种新的视频对象分割算法。利用人眼的视觉特点，即对运动（时间梯度）和边缘（空间梯度）都特别敏感，把帧间运动变化检测（时域定区间帧差累积）和图像的边缘检测结合起来，首先利用t显著性检验检测对称帧的帧间变化，再对检测出的初始运动变化区域进行时域定区间帧差累积计算，并进一步整合形成记忆掩膜（MT）；然后应用改进的Kirsch边缘检测算子较为精确地检测当前帧中所有的边缘信息，减少MT膜中的残留噪声，并通过时空滤波获得语义视频对象平面；最终选择性的应用填充及形态学处理操作，实现视频对象的分割。实验结果验证了本文算法的有效性和准确性。

英文摘要：

In order to solve the problems such as the inaccuracy of the segmentation contour extraction, occlusion, and irregular motion in video object segmentation methods, a novel video object segmentation method is proposed. Based on the human visual characteristics that human are sensitive to motion （tem- poral gradient） and edge （spatial gradient） especially, the inter frame motion change detection （the accu- mulation of fixed temporal interval frames difference） and image edge detection are combined to segment moving objects from stationary background precisely. First,t-distribution significance test is used to de- tect the inter frame changes of symmetrical frames;Second, the accumulation of fixed temporal interval frames difference of the detected initial motion change region is calculated,and then it can be integrated to form the movement memory template; Third, an improved Kirsch edge detection operator is used to detect all the edge information in current frame accurately;Fourth, spatiabtemporal filter is used to re- duce the residual noises in memory template and,extract the semantic video object plane~Fifth, the video objects segmentation can be obtained finally by applying filling and morphology operation selectively. By comparing our experimental results with other popular algorithms＇, the results indicate the validity and accuracy of the proposed algorithm.