基于模糊聚类的超像素图像分割算法

Superpixel Image Segmentation Based on Fuzzy Clustering

针对传统的模糊聚类算法在图像分割时出现的信息保留不足和抗噪性差等问题,提出一种基于模糊聚类的超像素图像分割算法。该算法首先在全局范围利用抑制式模糊c均值聚类算法确定像素的所属分类,然后在局部范围利用简单线性迭代聚类算法融合图像中物体的边界信息,最后通过计算超像素分割网格划分下的均值隶属度完成图像分割。算法既保留模糊聚类的粗模块分割优势,又利用超像素提高对边界信息的融合能力,达到优化分割效果的目的。在数值实验部分,分别在人工合成图像和Berkeley数据集等真实图像上检测算法效果,验证了边界分割能力和抗图像噪点能力。

Aiming at the problems of insufficient information retention and poor anti-noise properties of traditional fuzzy clustering algorithms in image segmentation,this paper proposes a super-pixel image segmentation based on fuzzy clustering.The segmentation method first uses the Suppressed Fuzzy C-Means clustering algorithm to determine the classification of pixels in the global range.The boundary information of the object in the image is then fused at the local scale by using the Simple Linear Iterative Clustering.Finally,the image segmentation is completed by calculating the mean degree of affiliation under the superpixel segmentation grid division.This method not only retains the advantages of fuzzy clustering in coarse module segmentation,but also uses superpixel algorithm to improve the fusion ability of boundary information,so as to achieve the purpose of optimizing the segmentation effect.In the numerical experiment section,the effect of the algorithm is tested on synthetic images and the Berkeley data set respectively,and the property of boundary segmentation and anti-image noise is verified.