【摘要】 模糊C均值算法(FCM)具有良好的聚类性能从而被广泛应用于图像分割领域,但其存在距离测度鲁棒性差、需预先给出初始聚类数目、未考虑图像局部相关特性等问题。本质上讲,FCM算法是一种局部搜索优化算法,如果初始值选择不当,不仅需要更多的迭代次数,而且会收敛到局部最优解。针对上述问题,结合进化聚类(ECM)和FCM算法,提出了一种遥感图像分割的新方法。利用ECM解决模糊C均值聚类算法的初始化中心选择问题,再利用FCM算法对获得的聚类中心进行优化,完成模糊聚类划分,通过去模糊化转换为确定性分类,实现聚类分割。实验结果表明,该方法能以较少的迭代次数收敛到全局最优解,具有较好的稳定性和鲁棒性,有较好的分割效果,提高了遥感图像分割方法的效率。更多还原

【Abstract】 Fuzzy C-means(FCM) algorithm has good clustering efficiency,for which it is widely used in the field of image segmentation.However,problems such as weak robustness of distance measure,the number of initial clustering to be given in advance and not considering local image feature still exist.In essence,FCM is a local search algorithm.Improper selection of initial value will lead to the need for more iterations and convergence to local optimal solution.In response to these problems,by combining evolving clustering(ECM) with FCM algorithm,put forward a new method of remote sensing image segmentation.By using FCM algorithm to solve the choice of ECM’s initialization clustering centers and using FCM to optimize the obtained centers,completed fuzzy clustering.And by converting fuzzy into a certainty classification,realized clustering segmentation.Experiment results show that the algorithm is typical of relatively less iterations of convergence to global optimum,good stability and robustness.It helps to produce better segmentation effect and improve the efficiency of remote sensing image segmentation.更多还原