基于分裂式K均值聚类的图像分割方法

模糊C均值聚类(FCM)算法是一种有效的无监督图像分割方法，适用于任意分类数，不需要预知图像特征，但其聚类效果直接受待分类样本噪声和分类初始条件的影响。因此，提出了一种适用于彩色图像分割的分裂式K均值聚类(FKM)算法，该算法首先使用中值滤波对分类样本去噪，然后使用一种分裂聚类法对图像样本进行预分类，得到一组样本集初始划分，最后以这组划分为起点，使用基于概率距离的K均值聚类对图像分割进行迭代优化。实验结果表明，该算法可以避免FCM的误分类，诸如陷于中心死区、中心重叠和局部极小值，而且提高了分割速度。

Image segmentation by fissive K-mean clustering method

Fuzzy C Means (FCM) clustering algorithm is an efficient unsupervised segmentation method, which is suitable for any classification number without the need to predict the image characteristics, but its clustering result has direct impact from sample noise component and the set of initial conditions. Therefore, a Fissive K-Means(FKM) clustering algorithm for color image segmentation was proposed, which firstly denoised the sample data using median filtering, then preclassified the image samples according to a fissive clustering method to get an initial partition of sample set, finally iteratively optimized segmentation using K-means clustering based on the rule of probability distance from the initial partition. The experimental results show that the algorithm can avoid the misclassification of FCM such as dead center, center overlapping and local minima, and accelerate the segmentation speed.