一种新的视频运动对象分割技术

Novel moving object segmentation technique for video sequences

本文将用于自然静态图像抠图领域的Knockout技术用于视频运动对象的精确分割,给出了一种新的视频运动对象分割方法。在利用积累差异技术自适应建立背景模型、采用背景差法初步提取当前帧视频对象区域的基础上,用本文提出的方法,自动标记原本采用Knockout技术进行抠图时,需手动进行标记的前景轮廓区域、背景轮廓区域及未知区域。在用本文方法完成对3个区域的自动标记后,再利用Knockout技术对当前帧的视频运动对象进行精确分割。本文还设计了一种新的变系数空域滤波器,该滤波器能有效地对背景差图像进行显著增强。同时,对Otsu自适应阈值化方法进行了改进,改进的方法能更准确地对背景差图像进行阈值化。

This paper applies Knockout technique to moving object segmentation, which is commonly used in image matting area, and a novel moving object segmentation technique is proposed. Based on the detection results obtained using accumulative differences, the foreground contour region, background contour region and unknown region are automatically marked, and then the Knockout method is applied to video object segmentation. A novel variable coefficient spatial filter is designed, which could effectively enhance the background difference image. Otsu method is improved, which could threshold the background difference image much more accurately. Experiment resultsshow that the proposed method can extract video object accurately and is also an effective method.