Sparse self-representation incorporated and residual driven adaptive fuzzy C-means clustering
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,VOC 2012 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.
NETL
NSTL
万方数据
