基于改进分水岭算法的粘连物体图像分割

Image Segmentation about the Touched Object Based on Improved Method of Watershed Algorithm

本文分析了分水岭算法应用在图像分割时存在的缺陷:对噪声和微弱边缘过于敏感;对其容易造成图像过分割(即生成多余的区域)的特点进行改善,进而提出了改进的分水岭算法。该算法首先使用传统分水岭算法原理对粘连物体图像进行粗略分割,然后通过模糊C均值聚类算法对过分割的图像进行聚类合并,完成精细分割。本文的实验结果证明:本方法不仅对粘连物体图像进行了有效地分割,而且克服了分水岭算法的过分割问题。

In this paper, the drawback of the watershed algorithm applied to image segmentation is analyzed. It is too sensitive to the noise and the weak edge. To improve the problem that the algorithm is easy to produce over-segmenta- tion results, i.e., generating excess area, an improved watershed algorithm was proposed. Firstly, traditional watershed algorithm was used to segment the touched object roughly. Then, the fuzzy C-means clustering algorithm was used for merging cluster to the over-segmentation image, completing the fine segmentation. Experimental results show that it not only completes image segmentation about the touched object, but also improves the over-segmentation of the watershed algorithm effectively.