一种鲁棒的无监督聚类图像分割算法

针对目前基于模糊C-均值聚类图像分割算法的噪声敏感问题, 提出一种基于无监督可能性聚类的自动加权图像分割算法. 该算法先应用均值漂移迭代确定可能性C-均值聚类算法的初始化中心, 利用可能性聚类的模式搜索性质自动确定聚类划分; 然后根据像素间灰度值关系进行图像加权, 通过将加权系数与像素噪声的可能性相关联, 降低噪声对图像分割的影响. 实验结果表明, 相对于基于模糊C-均值聚类的图像分割算法, 该算法不仅取得了较好的分割效果, 而且无监督分割时计算效率更高, 对噪声的鲁棒性更强.

A Robust Image Segmentation Algorithm Based on Unsupervised Clustering

Aiming at the problem that sensitivity to noise was evitable for image segmentation algorithm based on fuzzy C-means clustering, we proposed an image segmentation algorithm based on unsupervised possibilistic clustering with automatic weighting. Firstly, the initialization center of the possibilistic C-means clustering algorithm was determined by means of mean shift iteration, and cluster partition was automatically determined by pattern search property of possibilistic clustering. Then, the image was weighted according to the relationship of the gray values of the pixels. By correlating the weighted coefficients with the possibility of pixel noise, the influence of noise on image segmentation was reduced. The experimental results show that compared with the image segmentation algorithm based on fuzzy C-means clustering, the proposed algorithm not only achieves better segmentation result, but also has higher computational efficiency of unsupervised segmentation and more robust to noise.