以统计变化检测为基础的实时分割视频对象新方法

为了克服利用变化检测分割视频对象过程中的噪声、复杂运动、暴露背景的影响，提出了一种基于统计变化检测的实时分割视频对象新方法。在该方法中，由于统计变化检测技术是利用t分布来有效消除噪声的影响，而不需要估计噪声的方差，而且可利用间隔为k的两帧图像代替连续两帧来进行变化检测，因此可以很好地处理关节运动和慢运动；另外，对两个连续的统计变化检测结果取交集还可以消除暴露背景的影响，并能消除大部分的残留噪声，且几乎不增加计算量，因此统计变化检测可作为视频对象分割的基础，试验结果表明，该方法不仅解决了传统的变化检测过程中的噪声、复杂运动以及暴露背景影响，而且能够自动实时地分割视频对象，以满足MPEG-4等基于对象的视频应用。

A New Method for Real-time Segmenting Video Objects Based on Statistical Change Detection

In order to eliminate the effect caused by noise, complex motion, and uncovered background in the process of real time segmenting video objects, a new method for real time segmenting video objects based on statistical change detection is proposed. During the process of statistical change detection, because t distribution is used to eliminate the effect of noise, the noise variance is not needed. Two frames at a distance of k are used for statistical change detection instead of two successive frames, thus articulation motion and slow motion can be better coped with. The intersection between two successive statistical change detection results can eliminate the effect of uncovered background, and most of the residual noise are eliminated at the same time without adding any computation complexity. The experimental results show that the new method solves the problems existing in the process of traditional statistical change detection caused by noise, complex motion, and uncovered background, and it can automatically segment video objects by real time. It can meet the requirements of many object based video applications such as MPEG-4.