基于改进的直觉模糊核聚类的图像分割方法  

Image Segmentation Based on Intuitionistic Fuzzy Kernel c-Means Clustering Algorithms

作　　者：徐小来 房晓丽 XU Xiaolai;FANG Xiaoli(College of Electrical and Information Engineering, Hunan Institute of Information Technology, Changsha 410151, China)

机构地区：[1]湖南信息学院电子信息学院

出　　处：《计算机工程与应用》2019年第17期227-231,共5页Computer Engineering and Applications

基　　金：湖南省教育厅科学研究重点项目（No.18A511）

摘　　要：针对噪声图像模糊性的本质,提出了基于改进的直觉模糊核聚类的图像分割方法。采用直觉模糊集描述噪声图像包含的不确定性信息,将图像的灰度信息转换到直觉模糊域进行处理;将模糊核聚类拓展为直觉模糊核聚类,在图像的直觉模糊域进行聚类;通过高斯核函数和欧氏距离分别对像素8-邻域的灰度和空间信息进行建模,综合平衡灰度和空间信息对聚类的作用,并将其作为惩罚项加入到直觉模糊核聚类的目标函数中;通过梯度下降法,推导了迭代求解算法;通过典型的合成图像和自然图像分割实例,验证了所提算法的有效性和鲁棒性。To handle the uncertainty of noisy image, the image segmentation based on improved intuitionistic fuzzy kernel c-means clustering algorithms is proposed. Firstly, the intuitionistic fuzzy set is used to describe the uncertainty information of noisy image, and the gray values of image are transferred to intuitionistic fuzzy domain. Secondly, fuzzy kernel c-means clustering algorithms are extend to intuitionistic fuzzy kernel c-means clustering algorithms, and the image is clustered in intuitionistic fuzzy domain. Thirdly, the intuitionistic fuzzy factor which Gaussian kernel function and Euclidean distance are used to model the grey level and spatial information of 8-neighbor separately, is added into the object function of intuitionistic fuzzy kernel c-means clustering algorithms. Then, the iterative formulas are deduced by gradient descent methods. At last, experiments executed on one synthetic image and nature image demonstrate the effectiveness and robustness of the proposed method.

关 键 词：直觉模糊集 直觉模糊聚类 图像分割 核方法 模式识别 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]