一种基于Ｃ均值的模糊核聚类图像分割算法

模糊核聚类算法是一种结合无监督聚类和模糊集合概念的图像分割技术,已广泛应用于图像分割领域,但其算法对初值敏感,很大程度上依赖初始聚类中心的选择,并且容易收敛于局部极小值,用于图像分割时,隶属度的计算只考虑了图像中当前的像素探值,而未考虑邻域像素探间的相互关系,故对分割含有噪声图像不理想。故提出了一种改进的模糊核聚类图像分割算法,先通过数据约简,不损失数据聚类结构的前提下对数据进行挖掘,然后在模糊核聚类算法中引入特性核函数,将约简后的数据映射到高维非线性特征空间进行划分,最后再利用表征邻域像素的参数来修正当前空间像素的隶属度。实验结果表明,提出的算法较好地解决了模糊核聚类算法在局部极值处收敛和在迭代过程中出现停滞等问题,最终得到最佳全局聚类,迭代次数降低明显,并具有高鲁棒性、对噪声不敏感的特点。

A fuzzy kernel clustering image segmentation algorithm based on C-means

Fuzzy clustering algorithm is an image segmentation combining of unsupervised clustering and fuzzy set concept which is widely used in image segmentation. But fuzzy clustering algorithm is sensitive to initial value. It is largely dependent on the choice of the initial cluster centers and easy to converge to local minimum value. Only the pixel values of the current image are considered when calculating the degree of membership, instead of the relationship between the neighborhood pixels. So it is unsatisfactory for noised image segmentation. An image segmentation based on improved fuzzy clustering algorithm is presented in this paper. The data is briefed under the premise of not losing data clustering structure. And then nuclear mapping is introduced into linear space so that the briefed data is mapped to high-dimension feature space to be divided. Lastly, space neighborhood pixel values are used to correct the membership of current pixel space. The simulation shows that the algorithm presented in this paper effectively solves the problems like local optimum convergence and stagnation in the iteration process. The best global clustering is achieved with much lower iterations. The algorithm has advantages such as efficient robust and it is insensitive to noise.