基于分水岭区域差异性的多级图像分割

分水岭分割算法的不足之处在于过分割,即生成大量的小区域而使目标物淹没其中．文中通过定义分水岭区域的差异性给出一种多级图像分割方案,克服了传统的解决过分割问题时需要预先设置标记点的缺陷,也避免了多分辨率分割中边缘模糊和移位的问题．该算法可以应用于多种类型的图像。并经实验证明具有良好的分割效果．

A Hierarchical Image Segmentation Algorithm Using Dissimilarity Between Neighboring Watershed Regions

One negative effect of watershed image segmentation is over-segmentation, i.e., so many tiny regions are generated that objects of interest can not be identified easily. This paper presented a hierarchical watershed image segmentation scheme by utilizing the dissimilarity between neighboring watershed regions. There is no need to set markers as with those traditional ones to overcome over-segmentation. In addition, there is no edge blurring encountered in those multi-scale watershed segmentation with a series of Gaussian filters. Experiments with various types of images were given and its application to interactive segmentation was provided.