Abstract: In order to resolve the problems of sensitivity to noise and over-segmentation existing in traditional watershed algorithm，a new image segmentation method combined watershed translation and FCM clustering algorithm incorporating spatial information is proposed in this paper.Firstly，the median filtering is performed on the original image to simultaneously reduce noise and preserve edges.Then watershed translation is used for segmenting the image into many small regions，and a FCM clustering algorithm incorporating spatial information is proposed for merging.In the modified FCM clustering algorithm，a graph theory based metric incorporating region feature and spatial information is proposed.With the small regions as its vertices，a weighted graph is generated.The shortest paths between different vertices are calculated for measuring the distances between different regions.The experimental results show that，combining the region feature information and spatial information，the presented method gives more satisfying result than the traditional methods.

Key words: image segmentation, watershed, FCM, spatial information

摘要： 提出了一种分水岭变换和结合空间信息的FCM聚类相结合的图像分割方法。方法采用基于图论的结合区域特征信息和空间信息的距离度量，以分水岭变换得到的图像分割小区域为节点构建一个连通加权图，通过计算图上不同节点之间的最短路径来度量不同区域之间的相似程度，从而实现过分割小区域的合并。该方法综合考虑了区域的特征之间的差异和空间位置的差异，与传统的FCM聚类方法在特征空间进行聚类相比，具有较强的噪声抑制能力。图像分割的实验结果证明了该算法的可行性和有效性。

关键词: 图像分割, 分水岭变换, 模糊C均值聚类, 空间信息