提出一种时空融合的运动目标分割方法.在时域方面,采用时间轴一维小波变换提取运动对象,然后用独立成分分析法提取独立的运动对象,并基于灰度直方图进一步提取视频对象;在空域方面,提出对轮廓提取后的图像进行分水岭变换的改进方法.与COST211AM算法比较表明,文中方法能更完整、准确地提取出运动对象.
This paper addresses the problem of spatial-temporal segmentation of video sequences. In the temporal segmentation, we use the 1D wavelet transform to get motion information of adjacency frames, then the independent component analysis is used in removing noise and extracting independently moving objects. Furthermore, change detection algorithm based on histogram is adopted to segment moving objects. In the spatial segmentation, in order to alleviate over-segmentation problem, the inter-frame was partitioned into regions by an improved watershed algorithm, which processes the edge of image rather than image itself. Finally, our algorithm is compared with the famous COST211 AM. The experimental results show that our algorithm can extract moving object more accurately than the COST211 AM but with a better shape preserving ability.