【摘要】 针对模糊C均值聚类算法对初始值敏感、易陷入局部最优以及谱聚类算法无法处理样本量过大的问题,提出了一种将模糊C均值聚类算法与谱聚类算法相结合的模糊谱聚类算法应用于彩色图像分割;大致分为三步:第一步对图像进行预处理,将颜色空间由RGB空间转换为Lab空间;第二步对特征空间进行冗余模糊C均值聚类算法得到冗余类;第三步由冗余类的隶属度矩阵和聚类中心矩阵得到冗余类的特征空间,并根据贴进度和传递闭包将该特征空间转换为冗余类的相似度矩阵进行谱聚类,完成冗余类的合并;实验结果表明,与模糊C均值聚类算法相比,模糊谱聚类算法对于初始值敏感问题、易陷入局部最优以及只能识别团状的蔟得到了很好的解决,从而使彩色图像分割结果更加合理。更多还原

【Abstract】 Aiming at the fuzzy C-means clustering algorithm to the initial value sensitive,easy to fall into local optimum and spectral clustering algorithm cannot handle the sample volume is too large* a fuzzy C-means clustering algorithm and spectral clustering algorithm combining fuzzy spectral clustering algorithm is applied to color image segmentation is proposed.Roughly divided into three steps.The first step of image of pretreatment,the color space by the RGB color space conversion for lab space;the second step of feature space redundancy fuzzy C-means clustering algorithm to obtain the redundant j the third step by the redundancy class membership matrix and the cluster center matrix are redundant feature space and according to the closeness degree and transitive closure convert the feature space for redundancy of the similarity matrix for spectral clustering,merging redundant.Experimental results show that and fuzzy C-means clustering algorithm compared to fuzzy spectral clustering algorithm for the initial value sensitivity,is easy to fall into local optimal and can identify clusters of cocooning has been very good solution,so that the color image segmentation results are more reasonable.更多还原