关键词: 变化检测 视频对象 分割 时空滤波器

Abstract:

With the development of object-based video applications, the segmentation of video objects becomes one of the hotspots that people research. This paper presents a new method for the fast segmentation of video objects based on the change detection and spatial-temporal filters. It can detect changes between frames by the t-distribution significance test without the noise variance. We use two frames at a distance of k for change detection instead of two successive frames, thus the articulation motions and slow motions can be processed properly. The uncovered background due to the motions of video objects can be fast and effectively eliminated by spatial-temporal filters, and the residual noise in change detection masks can be reduced at the same time. Finally, the segmentation of video objects is implemented by morphology processing

Key words: change detection, video object, segmentation, spatial-temporal filter