基于变化检测的视频对象分割方案

Video object segmentation scheme based on change detection

由于运行时平均法建立的背景随更新系数的不同而波动,提出了一种新的更新系数选择方法,并利用了基于分块代替基于像素的帧差处理方法来减少差分技术对噪声的敏感性。在研究背景差分容易把背景中变化大的部分判决为运动物体,而帧间差分容易把运动物体中缓慢变化的部分判为背景的基础上,提出了一种新的帧间差分、背景差分和边缘检测相结合的视频对象分割方案。实验结果表明,新方案实时性高,分割得到的视频对象相对于单独使用帧间差分或背景差分的效果有明显的改进。

Taking into account the fact that background established by running-with-average method easily fluctuates with various updating coefficients,a new selection method of updating coefficient is proposed,and the block-based frame difference processing method instead of pixel-based method is employed to tackle the weakness of difference technique sensitive to noise.By referring to our investigation on the two disadvantages that background difference technique tends to decide large changing background into motion object and frame difference technique facilitates error judgment of gentle moving part of the motion object to background,a new video object segmentation scheme is introduced based on a combination of frame difference,background difference and edge detection techniques.The experimental results show the proposed scheme obtain high real-time advantage and good visual quality of segmented video object compared with that of the single frame difference or background difference techniques.