中文摘要：

视频分割是视频处理领域的基本问题,也是该领域的研究前沿和热点问题之一,在视频监控、编辑合成等方面都有着重要的应用。传统的视频分割方法大多依赖帧间局部相似性或运动的连续性进行区域划分,对遮挡、大幅度运动等情况的分割效果较差,需要大量的手工交互。论文通过在视频空间建立跨时空域的相似性邻接关系,提出一种新的视频分割图分割模型,并且采用最大流/最小割算法对相应的模型进行快速求解,从而实现视频的有效分割。论文算法只需要用户在视频的关键帧图像上进行少量交互,便自动获取整个视频分割结果;并且,该分割过程不受前景对象遮挡、快速运动等情况的影响,具有很好的稳定性。

英文摘要：

Video cutout is a fundamental yet frontier hot topic in video processing area,which plays an important role in video supervision,video editing,video composition and so on.Traditional methods mostly cut out the regions depending on the local similarity between the successive frames and the continuity of motion which are not able to accomplish a satisfying cutout results when the object regions are moving quickly or under occlusion.In this situation,lots of interaction may be needed.Based on the spatio-temporal similarity neighborhood,this paper presents a new model for efficient video cutout,which can be solved by max-flow/min-cut algorithm quickly.This method only needs very little interaction from the users,and more importantly,it is robust to the situations with foreground objects have fast motion or occlusion.