鲁棒总变分核空间模糊聚类分割算法研究  

Research on robust total generalized variation kernel space fuzzy clustering segmentation algorithm

作　　者：蔡秀梅 刘笑 吴成茂 CAI Xiumei;LIU Xiao;WU Chengmao(Xi’an University of Posts&Telecommunications,School of Automation,Xi’an 710121,China;Xi’an University of Posts&Telecommunications,School of Electronic Engineering,Xi’an 710121,China)

机构地区：[1]西安邮电大学自动化学院,西安710121 [2]西安邮电大学电子工程学院,西安710121

出　　处：《激光杂志》2023年第1期148-157,共10页Laser Journal

基　　金：陕西省教育厅科研计划项目资助(No.20JC32)。

摘　　要：针对现有的总变分模糊聚类分割算法在强噪声干扰图像分割中难以获得准确有效分割结果的不足,提出了鲁棒总变分核空间模糊聚类分割算法。该算法首先在鲁棒模糊聚类分割算法的目标函数中引入了总广义变分(TGV)正则化,消除图像中不需要的噪声和伪影;其次引入局部空间信息、局部灰度信息以及非均匀隶属函数从而构造一个新的模糊局部信息因子,在保证噪声抑制的同时保留图像中更多细节信息;最后将改进的聚类算法推广至核空间,使原空间线性不可分的像素样本点变成线性可分或近似线性可分,从而更好地给每个像素分配更高的隶属度。实验结果表明,与现有的总变分模糊聚类分割算法相比,建议算法在强高斯噪声干扰情况下的分割精度提高了14.7%,对强高斯噪声有较好的鲁棒性以及分割性能。Aiming at the shortcomings of the existing total variational fuzzy clustering algorithm that it is difficult to obtain accurate and effective segmentation results in strong noise interference image segmentation,this paper proposes a robust total variational kernel spatial fuzzy clustering algorithm.Firstly,the algorithm introduces total generalized variational(TGV)regularization in the objective function of the robust fuzzy clustering segmentation algorithm,eliminating unwanted noise and artifacts in the image;Secondly,local spatial information,local grayscale information,and non-uniform membership function are introduced to construct a new fuzzy local information factor,which retains more detailed information in the image while ensuring noise suppression;Finally,the improved clustering algorithm is generalized to the kernel space,so that the original spatial linearly indivisible pixel sample points become linearly separable or approximately linearly separable,to better assign higher membership to each pixel.Experimental results show that compared with the existing total variational fuzzy clustering and segmentation algorithm,the segmentation accuracy of the proposed algorithm under the interference of strong Gaussian noise is improved by 14.7%,which has good robustness and segmentation performance for strong gaussian noise.

关 键 词：图像分割 模糊聚类 局部空间信息 局部灰度信息 模糊局部信息因子 

分 类 号：TN911[电子电信—通信与信息系统]