基于全散度的自适应鲁棒图形模糊聚类算法

针对图形模糊聚类对灰度分布不均匀及噪声干扰图像无法获得满意分割结果的不足，提出一种基于全散度的自适应鲁棒图形模糊聚类分割算法。全散度和像素邻域信息相结合，得到一种改进的全散度；改进的全散度引入图形模糊聚类最优化模型，并嵌入像素空间邻域信息。当前聚类像素与邻域像素均值的偏差作为该鲁棒聚类分割模型的正则因子，促使该聚类对强弱噪声具有自适应抑制能力。测试结果表明，与现有的图形模糊聚类、鲁棒模糊聚类等算法相比，自适应鲁棒全散度图形模糊聚类分割算法的分割效果和抗噪鲁棒性均有明显改善。

Adaptive Robust Picture Fuzzy Clustering Algorithm Based on Total Bregman Divergence

As picture fuzzy clustering algorithm is not suitable for segmentation of image with noise or inhomogeneous intensity, an adaptive robust picture fuzzy clustering segmentation algorithm based on total Bregman divergence is proposed. An improved total Bregman divergence is constructed by combination of existing total Bregman divergence and neighborhood information of image pixel, which is suitable for image segmentation. It was introduced into the picture fuzzy c-means clustering optimization model, and a robust total Bregman divergence-based picture fuzzy clustering algorithm, in which the pixel spatial neighborhood information was embedded, was obtained. The difference between the gray values of current clustering pixel and its neighborhood pixel is used as the regularization factor of the robust picture fuzzy clustering model based total Bregman divergence, and thus the robust clustering segmentation method would be capable of suppressing the noise adaptively. The results show that the segmentation quality and anti-noise robustness of the proposed segmentation algorithm are improved more significantly than those of the existing picture fuzzy clustering and other robust fuzzy clustering algorithms.