Research On Adaptive Regularization Weighted Robust Fuzzy Clustering

As the common foundation and premise of image understanding and machine vision,image segmentation has an extremely important research value in image processing.At present,it has been playing an increasingly important role in human life and work,such as biomedical engineering,intelligent transportation,intelligent agriculture,earth observation and remote sensing and other fields.However,due to the continuous improvement of the quality of image segmentation and the inherent fuzziness of images,the traditional image segmentation technology can no longer satisfy the demand of human life and work,so it is imperative to further study image segmentation technology.In view of the complexity,randomness and uncertainty of the image itself,fuzzy theory can be used to describe the fuzziness and uncertainty for the image in a more delicate and appropriate way.Therefore,fuzzy sets provide a new and more powerful technical support for the development of image segmentation theory system.Fuzzy c-means clustering(FCM)algorithm has attracted the attention of scholars at home and abroad because of its advantages of simplicity and efficiency.However,it adopts a single membership degree to express the attribution problem among samples,which can not fully describe the uncertain information among samples,resulting in some deviation in sample classification.The FCM regards the samples as isolated sample points,and ignores the connection between samples,which leads to the phenomenon of misclassification of singular values in samples,resulting in poor robustness of the algorithm.Aimed at the above limitations,many scholars have made a series of improvements to FCM.For example,in view of the limitation of single membership degree,the picture fuzzy clustering algorithm was proposed by combining picture fuzzy set with FCM,and in view of the lack of spatial information,a series of FCM algorithms based on spatial information were proposed by combining neighborhood information with FCM,which greatly improved the segmentation performance of the algorithm.On this basis,this paper adopts the spatial information,KL divergence and possibility theory to improve and optimize the fuzzy clustering algorithm,mainly focusing on the following algorithms:1.To improve the accuracy and robustness of existing picture fuzzy clustering and solve the problem of parameter selection of spatial constrained regularization,an adaptive weighted picture fuzzy clustering algorithm is proposed.Firstly,a new symmetric quasi quadratic regularization term is used to solve the time-consuming problem of exponential regularization in existing picture fuzzy clustering.Secondly,considering the correlation between the gray level of the current pixel and its neighboring pixels,the adaptive weight is introduced to fuse the current pixel and neighborhood mean,and the weight entropy constraint is embedded into the clustering objective function to solve the problem of parameter selection.Finally,the local spatial information constraint term of current pixel isconstructed by using the picture fuzzy partition information of neighborhood pixel,and the picture fuzzy partition information of current pixel is modified by using this constraint term to modify the clustering center obtained by iteration.The segmentation results show that the improved algorithm has potential advantages in segmentation accuracy and anti-noise robustness.2.In view of the fact that the variants of FCM cannot realize the adaptive adjustment of spatial information constraints,an adaptive weighted robust fuzzy clustering algorithm based on KL divergence is proposed.Firstly,the neighborhood pixel mean and median information are used to assist the current pixel clustering,and the adaptive weight coefficient is introduced and the adaptive weight entropy regularized term is added to construct the objective function of fusing spatial neighborhood mean and median information.Secondly,considering the relationship of membership degree of neighborhood pixels,the KL divergence and spatial constraints of membership degree are combined to constrain the membership degree within the cluster,so as to further improve the segmentation accuracy of the algorithm.Finally,the tests of synthetic image,medical image and remote sensing image are carried out respectively,and results show that the proposed algorithm possesses stronger universality for noise and more satisfactory segmentation effect for noisy image than other comparison algorithms.3.In order to further improve the performance of the algorithm,a new adaptive robust fuzzy clustering algorithm based on the the KL divergence and possibility theory is proposed.Considering that the basic idea of possibilistic c-means(PCM)algorithm is to relax the normalization constraint of membership degree,this relaxation constraint idea is introduced into the above-mentioned adaptive weighted robust fuzzy clustering algorithm,and a new adaptive possible weighted robust fuzzy clustering model with the sum of weighted coefficients not equal to one is constructs.The proposed algorithm further enhances the noise suppression ability of the algorithm.