融合稀疏自表示和残差驱动的自适应模糊C均值聚类  

Sparse self-representation incorporated and residual driven adaptive fuzzy C-means clustering

作　　者：宋燕[1] 李元昊 李明[1] SONG Yan;LI Yuan-hao;LI Ming(Department of Control Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学控制工程系,上海200093

出　　处：《控制与决策》2024年第4期1333-1341,共9页Control and Decision

基　　金：国家自然科学基金项目(62073223);上海市自然科学基金项目(22ZR1443400);航天飞行动力学技术国防科技重点实验室开放课题项目(6142210200304)。

摘　　要：提出一种融合稀疏自表示和残差驱动的自适应模糊C均值聚类算法(R2AFCM).该算法的优点主要体现在以下两个方面:1)利用稀疏自表示技术求解样本数据的字典矩阵,并将其表征的全局信息考虑到目标函数中,充分考虑数据分布特点,改进传统模糊C均值聚类算法重点关注局部信息的不足;2)在目标函数中引入加权残差估计正则化项,与自适应模糊聚类算法的正则化项相结合,约束模型训练,有效降低混合噪声对分割结果的影响.在磁共振成像、VOC 2012数据集以及自然图像上进行对比实验,结果表明,所提出的聚类算法在添加了20%椒盐噪声和均值为0.4、方差为0.01的高斯噪声,以及50%椒盐噪声和均值为0、方差为0.1的混合噪声下与其他算法相比,具有更高的分割精度和更强的鲁棒性.This paper proposes an adaptive fuzzy C-means clustering algorithm that combines sparse self-representation and residual drive(R2AFCM).The advantages of this algorithm are mainly reflected in the following two aspects:1)Using sparse self-representation technology to solve the dictionary matrix of sample data,and taking the global information represented by it into the objective function,fully considering the data distribution characteristics,the lack of the traditional fuzzy C-means clustering algorithm focusing on local information is improved.2)The weighted residual estimation regularization term is introduced in the objective function,combined with the regularization term of the adaptive fuzzy clustering algorithm,which constrains the model training,and effectively reduces the influence of mixed noise on the segmentation results.Finally,comparative experiments are carried out on magnetic resonance imaging,VOC2012 dataset and natural images.The experimental results show that the proposed clustering algorithm adds 20% salt and pepper noise,with a mean value of 0.4 and a variance of Gaussian noise of 0.01,and 50% salt and pepper noise and mixed noise with mean value of 0 and variance of 0.1,it has higher segmentation accuracy and stronger robustness than other algorithms.

关 键 词：图像分割 自适应模糊聚类 混合噪声 稀疏自表示 加权残差估计 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]