快速自适应非局部空间加权与隶属度连接的模糊C-均值噪声图像分割算法

针对传统模糊C-均值聚类(FCM)算法难以对噪声图像进行分割的问题，该文提出一种快速自适应非局部空间加权与隶属度连接的模糊FCM抗噪图像分割算法。首先，利用一种非局部空间信息快速计算方法，将以图像所有像素为循环的原始非局部信息计算方法，改为以搜索窗口尺寸为循环，利用空间位移图像与递归高斯滤波的计算方法，克服非局部空间信息计算复杂的问题；其次，计算原始图像与非局部信息项的差值的平方，将其作为非局部信息项的自适应权重，并将差值的平方作倒数变换，作为原始图像的自适应权重；最后，将每个聚类簇中所有像素隶属度之和的对数平方加入目标函数的分母，形成隶属度连接，减少目标函数迭代次数。含噪人工与自然图像分割实验表明，该算法在分割准确度、平均交并比、归一化互信息、运行时间与迭代次数等性能方面优于其他几种FCM算法。

Fuzzy C-Means Clustering with Fast and Adaptive Non-local Spatial Constraint and Membership Linking for Noise Image Segmentation

Considering the problem of the low anti-noise performance when Fuzzy C-Means clustering (FCM) algorithm is applied to image segmentation, a FCM clustering algorithm with fast and adaptive non-local spatial constraint and membership linking is proposed in this paper. Firstly, in order to increase the computing speed of non-local spatial term, a fast method is proposed by modifying the loop based on all pixels in an image into a loop based on search window and by utilising spatial shift image and recursive Gaussian filter. Next, the squared difference between original image and non-local spatial term is calculated as adaptive weight of non-local information term. The squared difference is reciprocally transformed as adaptive weight of the original image. Finally, the membership linking is established to reduce the iteration steps before convergence by adding the square of the sum of all the membership degrees in every cluster in logarithmic form as the denominator of the objectvie function. Experiments on noisy artificial and natural images prove that this proposed algorithm has superior performance in terms of Segmentation accuracy, mean intersection over union, normalized mutual information, running time and iteration steps.