基于变化检测与时空滤波器的视频对象分割方法

Fast Segmentation of Video Objects Based on Change Detection and Spatial-Temporal Filters

随着基于对象视频应用的发展,视频对象的分割成为人们研究的热点。本文提出了一种结合变化检测与时空滤波器快速分割视频对象的新方法,该方法利用t分布显著性检验检测帧间的变化,不需要知道噪声的方差;利用间隔为k的两帧图像代替连续两帧进行变化检测,可以很好地处理关节运动和慢运动。然后通过时空滤波器快速有效地消除由于视频对象运动而露出的背景区域,同时能够减少变化检测掩膜中的残留噪声。最后,通过形态学处理实现视频对象的分割。

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