基于共有边界长度的改进分水岭算法

Modified Watershed Based on Recursive Merging

文章提出了一种新的改进分水岭算法,以克服传统的分水岭算法造成的图像过分割问题。改进的算法在传统的分水岭算法分割结果的基础上,按照面积由小到大的顺序,循环合并分割产生的区域。合并过程中,首先提出了一种新的邻域系统,用于计算需要合并的区域与相邻区域的共有边界长度,然后根据边界的长度确定与其最相似的区域,进而合并这两个区域,最后采用后处理算法来进一步细化区域合并后的分割结果。实验结果显示,该算法能很好地解决过分割的问题,使分割结果具有更准确的语义信息。

A novel modified watershed algorithm is proposed in this paper to reduce the over-segmentation problem produced by the classical watershed scheme.According to the areas of the regions in the segmented image,the small regions are merged recursively.In region merge process,a new neighborhood system is firstly proposed to calculate the length of edges between the region to be merged and its neighboring regions,then the best match region is determined according to the length of edges.After region merge,post-processing is applied to refine the segmentation results.Experimental results show that the proposed algorithm could overcome the problem of over-segmentation,and obtain more meaningful regions.