一种应用形态学滤波的视频对象时空分割算法

从视频图像中提取视频对象是基于内容的视频编码中的一项关键技术。本文提出了一种基于帧间运动信息和形态学滤波的视频对象时空分割算法。该算法首先利用分块高阶统计算法和基于最大类间方差的阈值算法得到目标的运动区域检测模板。然后,用基于交变序列重建滤波的分水岭算法得到前景对象的精确边缘。最后,用区域基时空融合方法将运动检测和形态学分割结果结合起来提取出视频对象。实验结果表明,本文算法能避免区域合并有效提取出具有精确边缘的视频对象,主客观分割效果理想。

Video object segmentation by morphological filtering based on temporal and spatial information

Video object extraction is a key technology in content-based video coding. A novel video object extraction algorithm is proposed in this paper. This algorithm is based on the temporal motion information and spatial morphological filtering. First, block higher order statistics and threshold via local maximal between-cluster variance are used to get the motion detection mask. Then a modified watershed algorithm based on alternating sequential filtering by reconstruction is applied to get the accurate edges. Finally, the video object with satisfactory boundary is obtained by fusing the temporal and spatial segmenting results. Experimental results show that the proposed algorithm can effectively extract the video object with relatively fine boundary while avoiding the complicated area merging, which is satisfying subjectively and objectively.