基于时空信息的自动视频对象分割算法

提出一种在通用视频序列中从背景里分离出运动对象的方法.首先,使用全局运动估计和补偿、场景变化检测技术进行预处理.然后,使用四阶统计显著性检测方法从帧差图像中分离出前景和背景,并使用连通成分标记算法和一连串形态学开启、闭合操作得到修正后的二值模板图.接着,使用对称差分技术消除覆盖显露的背景以及部分噪声.最后,使用模板匹配和更新,不仅能够得到快速变化的对象,而且能够得到视频对象暂时停止运动的部分.

Automatic video object segmentation algorithm based on spatio-temporal information

A segmentation method to separate the moving objects from the background in a generic video sequence is proposed. Firstly, the global motion estimation and compensation and the scene-change detection are achieved. Secondly, a robust discrimination between background and foreground by means of a higher-order statistical significance test is performed on a group of inter-frame differences, and a binary segmentation map is produced by means of a connected component labeling and a cascade of morphological filters. Thirdly, the symmetrical difference method is used to eliminate the effect of uncovered and covered background. At last,the mask matching and updating enable the algorithm to handle both objects that change quickly and objects that temporarily stop moving.