针对传统模糊C均值(FCM)聚类算法聚类数目难以确定,迭代速度慢,易陷入局部最优以及对聚类中心初始值的设置敏感等问题,提出一种融合全局最好和声搜索模糊C均值(GBHS-FCM)聚类算法。首先,利用全局最好和声搜索(GBHS)算法的全局性和鲁棒性的优点,得到初始聚类中心和聚类个数,再将其作为传统FCM聚类算法的初始聚类中心和聚类个数;其次,提出一种新颖的模糊聚类目标函数,将图像像素点邻域依赖特性考虑进来,与像素点灰度信息共同作用,增强了分割结果空间的连续性;此外,还采用了一种新颖的距离公式代替欧氏距离公式,增强了新算法对噪声的鲁棒性。仿真结果表明,新算法有效避免了传统FCM算法因初始聚类中心设置敏感而收敛到局部最优解,在聚类精度、速度和鲁棒性上均比传统FCM算法有所提高,针对具有不同特征的图像分割取得了较好的结果。

To overcome the shortcomings of the traditional FCM algorithm, such as the difficulty to determine cluster numbers slow iteration, and the tendency to plunge into local optimization,as well as sensitivity to, the initial values,an improved fuzzy c-means clustering combining with the global best harmony search algorithm(GBHS-FCM) is proposed. First, the initial cluster numbers and cluster centers of the FCM algorithm are obtained by the GBHS algorithm, while taking the advantages of global superiority and robustness of the GBHS algorithm. Then a new fuzzy clustering function is presented by combining the pixel intensity information and the spatial dependence to the neighboring pixels together, which enhances the spatial continuity of the segmentation results. Finally, a new distance formula is proposed to replace the traditional Euclidean distance formula, which enhances the robustness of the new algorithm to noise. The simulation results show that the GBHS-FCM algorithm performs better than FCM algorithm in accuracy, speed and robustness.