Abstract: Watershed transformation is a powerful morphological tool for image segmentation which can automatically generate a series of closed segmentation regions.However，the watershed transformation might give rise to over segmentation and it sensitive to noise.In order to overcome the inherent drawback of watershed algorithm-over-segmentation，a new method Hybrid Watershed and Improved FCM（HWIF） is proposed，which uses non-linear filter algorithm and a modified FCM algorithm to improve initial segmentation result obtained by the watershed transformation.This method uses information of regions gray and information between regions sufficiently compared with meanshift and region-merge method.Experiment result shows the proposed algorithm overcome the watershed algorithm-over-segmentation efficiently and obtain good segmentation.

Key words: Peer Group Filtering（PGF）, watershed, Fuzzy C-Means（FCM）, feature divergence

摘要： 分水岭变换是图像分割的一种强有力的形态工具，能够自动生成一系列封闭分割区域。其不足之处是过分割、对噪声敏感。为克服分水岭变换固有的缺点，综合利用非线性滤波和改进的FCM算法优化分水岭变换得出的初始分割，提出了一种新的混合分割算法-HWIF（Hybrid Watershed and Improved FCM）分割法。与MeanShift算法及区域合并算法相比，该方法充分利用了区域的灰度和区域间的空间信息。实验结果表明该算法能有效克服分水岭算法的过分割问题，且分割效果优于以上两种方法。

关键词: PGF滤波算法, 分水岭, 模糊C均值, 特征散度