摘要
针对聚类算法用于图像分割时造成的过度分割问题,提出一种带有深度邻域信息的模糊C均值聚类算法(FCM_DN).在传统的带有邻域信息的模糊C均值聚类算法基础上,引入类似高斯滤波的权重来表示像素点的位置差异,同时还引入中心点像素值与邻域点像素值之间的差异.相比于只考虑带有像素点之间位置差异的模糊C均值聚类算法,所提出的算法在聚类时可以使用更大的邻域,从而解决过度分割的问题.结果表明:在人造数据集上,本算法对于椒盐噪声和高斯噪声都有较强的鲁棒性;在现实数据上,本算法相比于11个先进算法在四个指标上有两个指标的表现都位于前三;在SED数据集的归一化互信息(NMI)结果对比中,本算法比其他算法高出1.78%~26.90%.
To solve the over-segmentation problem resulting from using clustering algorithm to segment pictures,a fuzzy C-means clustering algorithm with deep neighborhood information(FCM_DN) was proposed. On the basis of conventional fuzzy C-means clustering algorithm with neighborhood information,the weight similar to Gaussian filtering was introduced to represent the spatial difference between pixels points,and the difference between center pixel value and neighborhood pixel value was also introduced.Compared with fuzzy C-means clustering algorithms that only concern spatial difference between pixels points,the proposed algorithm could use larger neighborhood when clustering,so the over-segmentation could be solved.Results show that the proposed algorithm has strong robustness to salt&pepper noise and Gaussian noise on artificial dataset. On real dataset,the performance of the proposed algorithm is top three on two of four indexes compared to other eleven state-of-the-art algorithms.The performance on normalized mutual information(NMI) of the proposed algorithm is 1.78%~26.90% higher than that of other algorithms on SED dataset.
作者
车杭骏
陈科屹
王雅娣
刘晓阳
CHE Hangjun;CHEN Keyi;WANG Yadi;LIU Xiaoyang(School of Electronic Information Engineering,Southwest University,Chongqing 400715,China;Chongqing Key Laboratory of Nonlinear Circuits and Intelligent Information Processing,Southwest University,Chongqing 400715,China;Henan Province Big Data Key Laboratory of Analysis and Processing,Kaifeng 475004,Henan China;School of Computer and Information Engineering,Henan University,Kaifeng 475004,Henan China;School of Systems Engineering,National University of Defense Technology,Changsha 410073,China)
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2022年第11期135-141,共7页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(62003281,62103428,62106066)
中央高校基本科研业务费专项资金资助项目(SWU020006)
重庆市自然科学基金资助项目(cstc2021jcyj-msxmX1169)
湖南省自然科学基金资助项目(2021JJ40702)
河南省高等学校重点科研项目(22A520019)。
关键词
过度分割
模糊C均值
深度邻域信息
像素值差异
位置差异
over-segmentation
fuzzy C-means
deep neighborhood information
pixel value difference
spatial difference
