微分进化自适应模糊C均值分割算法  

Differential evolution of adaptive fuzzy C-means segmentation algorithm

作　　者：张志禹[1] 王彩虹[1] 张一帆[1] 

机构地区：[1]西安理工大学自动化与信息工程学院,西安710048

出　　处：《计算机工程与应用》2017年第23期135-141,共7页Computer Engineering and Applications

基　　金：国家自然科学基金重大项目课题四(No.41390454)

摘　　要：模糊C均值(FCM)聚类算法分割图像时,对图像的背景噪声和聚类算法的初始值比较敏感,为了克服这个问题,进而提出了微分进化模糊C均值分割算法。为了避免陷入局部极值,首先使用FCM聚类初始化,接着用改进的FCM进行模糊聚类;然后进行初始化种群操作,设置微分进化DE算法的参数,计算种群中每个个体的适应值,最后对满足条件的适应值进行变异、交叉、选择操作。利用DE算法的全局搜索优化能力,有效抑制了局部极值的产生和图像的背景噪声、纹理细节对图像分割效果的影响。还克服了对初值选择敏感的问题,保证图像分割边界的完整性,是一个比较高效的方法,有效地提升了分割效果。DE算法本身具有简单,快速,鲁棒性好等优点,利用这些优点可以有效地克服FCM算法的缺点。Fuzzy C Mean(FCM)clustering algorithm is sensitive to the initial values of the background noise and the clustering algorithm. In order to overcome this problem, a differential evolution fuzzy C means segmentation algorithm is proposed. To avoid falling into local extremum, firstly, fuzzy C means clustering is adopted to initialize, and the improved FCM is applied to carry out fuzzy clustering. In initialization population operation, the parameters of differential evolution algorithm are set, and the fitness value of each individual is calculated in the population, finally, conditions fitness value is studied to meet variation conditions, crossover and selection operation. The global search optimization ability of differential evolution algorithm can effectively restrain the effect of the background noise and texture details on the image segmentation. It also overcomes the problem that the initial value selection is sensitive to image segmentation, and it is a more efficient method to improve segmentation results. Conclusion differential evolution algorithm has the advantages of simpleness, fastness and robustness, which can effectively overcome the disadvantages of the fuzzy C means algorithm.

关 键 词：模糊C均值 图像分割 微分进化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]