滑动窗口式三维Watershed算法实现视频序列分割

视频序列分割是实现视频对象提取、处理和识别的基础,也是基于内容的视频压缩和检索的前提。目前普遍采用的watershed算法是全局涨水算法,用于视频分割时占用的存储和处理空间巨大。针对这一问题本文提出了新的滑动窗口式三维watershed算法(SW3DW),通过时间轴上的局部窗口对视频序列进行观察和分割,而在相邻窗口间采用水源区标号传递和局部流域更新相结合的方法,以较小的存储和处理空间实现了局域和全局处理的一致性。理论分析和实验证明该算法能有效性地实现视频序列的分割,为后续视频对象提取和处理创造了条件。

Sliding-Window 3-Dimensional Watershed Algorithm for the Segmentation of Video

The segmentation of video is the basis of the video object extraction, processing and recognition. It is also the premise of the content-based video compression and retrieving. In general, the widely used watershed algorithm is a global flooding method. For the segmentation of video, it requires large memory for the storage and processing. In this paper the sliding-window 3-dimensional watershed algorithm (SW3DW) has been put forward. Through a window along time axis, the watching and segmentation can be completed for this local video. By means of propagation of the water source label and updating of the local area between neighboring windows, it only uses the small storage and processing memory, and the result is in accordance with the global one. The theoretical analysis and experimental results have demonstrated that the segmentation of video can be efficiently implemented through this algorithm and it will benefit the video object extraction and processing in the following application.