基于模拟退火和粒子群改进的图像分割FCM方法

2018-04-04 09:10:16陆振宇夏志巍卢亚敏黄现云
现代电子技术 2018年7期

陆振宇 夏志巍 卢亚敏 黄现云

摘 要: 针对模糊C均值聚类(FCM)算法在分割图像时需要事先给出聚类数和容易陷入局部极小值的问题,提出一种新的FCM算法。首先,利用粒子群算法更新FCM的聚类中心,以加强算法的搜索能力,提高收敛速度;其次,根据模拟退火准则决定是否接受新的聚类中心,以得到当前迭代下的全局最优值;最后,设定有效性函数寻找图像的最佳聚类数,使算法具有自适应判断图像类别个数的能力。实验结果表明,该算法具有较好的全局收敛性,并且在未知聚类数的情况下能自适应寻找图像的最佳分类个数。

关键词: 自适应图像分割; 模拟退火算法; 粒子群算法; 模糊C均值; 聚类中心; 全局最优

中图分类号: TN911.73?34 文献标识码: A 文章编号: 1004?373X(2018)07?0036?05

Improved FCM method for image segmentation based on simulated

annealing and particle swarm optimization

LU Zhenyu1, 2, XIA Zhiwei1, LU Yamin1, HUANG Xianyun1

(1. School of Electronic & Information Engineering, Nanjing University of Information Science and Technology, Nanjing 210044, China;

2. Jiangsu Collaborative Innovation Center on Atmospheric Environment and Equipment, Nanjing 210044, China)

Abstract: Since the traditional fuzzy C?means (FCM) algorithm needs to give the clustering numbers in advance and is easy to fall into local minimum for image segmentation, a novel FCM algorithm based on simulated annealing algorithm and particle swarm optimization (PSO) is proposed. The PSO algorithm is applied to update the clustering center of FCM to enhance the search ability and convergence rate of the algorithm. And then the simulated annealing rule is used to decide whether to accept the new clustering center or not, so as to obtain the global optimal value of current iteration. The validity function is set to find the optimal clustering numbers of the image, and make the proposed algorithm have the ability to adaptively judge the numbers of an image category. The experiment results demonstrate that the algorithm has perfect global convergence, and is able to adaptively find the optimum catelory numbers of the image in the case of the unknown clustering numbers.

Keywords: adaptive image segmentation; simulated annealing algorithm; particle swarm optimization; fuzzy C?means; clustering center; global optimization

0 引 言

图像分割是根据一定的准则把图像分割成几个互不重叠且各具特色的区域。目前图像分割方法主要有:基于阈值的分割方法,如最大类间方差的自适应阈值分割[1]、基于粒子群优化的多级阈值分割[2];基于边缘检测的分割方法,如拉普拉斯算子[3]、结构化森林[4];基于区域生长和分裂合并的分割方法,如区域生长与神经网络相结合的方法[5]、水平集与区域生长相结合的方法[6];结合特定理论的分割方法,如基于分水岭的标号法[7]、深度卷积神经网络方法[8]。

聚类分割方法属于区域分割方法,它是一种无监督的统计方法,在图像分割领域应用非常广泛。FCM分割方法具有程序实现简单、不需要人为干预等方面的优势,但传统的FCM算法在求解目标函数时采用最速下降法,易使得迭代過程陷入局部最优解。……

登录APP查看全文