| As an important content in image processing technology,image segmentation can distinguish the content of different attributes in the image,so as to extract the object of interest.Because of the natural clustering of images in its feature space,image segmentation based on fuzzy clustering theory has been developed rapidly,and has been widely used in remote sensing,medicine,agriculture and other fields.However,the uncertainty of the image itself and the interference of noise in the transmission process greatly affect the effective image segmentation,resulting in unsatisfactory segmentation results.Therefore,it is of great theoretical value and practical significance to research image segmentation algorithms and find high accuracy and strong robustness of fuzzy clustering image segmentation methods.The traditional type-1 fuzzy C-means clustering algorithm realizes the automatic classification of sample data through iteration,which has the advantages of low storage cost and high efficiency,and has become one of the most widely used clustering algorithms.With the continuous improvement and development of fuzzy set theory,type-1 fuzzy set cannot fully describe the uncertainty in things effectively,while type-2 fuzzy set shows obvious advantages in dealing with higher-order uncertainty.The clustering method based on interval type-2 fuzzy set can segment the image effectively.Although existing interval type-2 fuzzy clustering algorithms can describe the higher-order uncertainty in the image more fully,it also has some shortcomings such as sensitivity to noise and difficult to select parameters reasonably.Therefore,in order to enhance the anti-noise performance of existing type-2fuzzy clustering algorithms and improve the segmentation accuracy of different types of images,this paper systematically studies and makes a series of improvements to interval type-2 fuzzy clustering algorithms.The main research contents are as follows:(1)In order to enhance the ability to suppress noise in existing interval type-2 fuzzy Cmeans related algorithms,and improve the segmentation performance of algorithm under noise case.Based on interval type-2 fuzzy C-means,this paper introduces rich neighborhood information and proposes a novel interval type-2 fuzzy clustering algorithm driven by deep structure information.Firstly,this paper applies neighborhood information to the iterative calculation of fuzzy membership,and uses the constraint relationship between pixels to reduce the impact of noise on pixel clustering,Secondly,the Gaussian kernel-induced distance metric is introduced to replace traditional square Euclidean distance in fuzzy clustering algorithms,which enhances the performance of the algorithm in processing highdimensional data.Finally,the adaptive weighted distance was constructed by the masterslave neighborhood information to update upper and lower fuzzy membership,which reduces the high computational complexity caused by selecting different fuzzy weighting exponents in interval type-2 fuzzy C-means related algorithms.Through a large number of image segmentation experiments with strong noise interference,the proposed algorithm has good segmentation performance and strong anti-noise robustness,which shows that the algorithm has certain feasibility and progressiveness.(2)Possibilistic C-means clustering algorithm cancels the constraint that the sum of membership is 1,which can more accurately describe the real properties of the sample,and shows stronger robustness than fuzzy C-means for clustering of noisy data sets.In order to further enhance the segmentation performance of type-2 fuzzy clustering algorithms on high noise images and improve the anti-noise robustness,this paper proposes a novel robust interval type-2 possibilistic fuzzy C-means clustering algorithm based on kernel function.Firstly,the possibility theory is combined with interval type-2 fuzzy C-means algorithm to describe uncertainty by using fuzzy membership and possibilistic typicality.Secondly,while introducing neighborhood information of pixels,a new local fuzzy factor is constructed by using structural similarity function between master and slave neighborhood to realize the fuzzy division of pixels.Finally,the interval fuzzy uncertainty is updated by adaptive weighted distance constructed by master-slave neighborhood information.By testing different types of noise images,it is confirmed that the proposed algorithm has obvious advantages in the case of high noise interference.It can not only effectively suppress the noise in images,but also reasonably retain the edges and details of original image.(3)The possibilistic construction of possibilistic fuzzy C-means clustering algorithm results in the "cluster consistency" of the algorithm in noise case,which can not effectively cluster data.In order to solve the overlapping problem of clustering centers in possibilistic clustering algorithms and improve the convergence of interval type-2 fuzzy clustering algorithms,this paper proposes a new enhanced adaptive interval type-2 kernel possibilistic fuzzy clustering algorithm.Firstly,based on the possibility theory,the construction framework of possibilistic fuzzy C-means clustering algorithm is modified,and the exponential updating method is used to improve the convergence of algorithm.Secondly,the interval fuzzy uncertainty is updated by adaptive weighted distance constructed by masterslave neighborhood information.Finally,combined with KM algorithm to realize type reduction and defuzzification.Compared with the traditional possibilistic fuzzy clustering algorithms,the proposed algorithm can obtain better segmentation results in noise cases,and shows strong anti-noise performance.(4)In order to improve the effectiveness of type-2 fuzzy clustering algorithm in dealing with different types of noise images,this paper combines unsupervised possibilistic clustering model and describes uncertainty by using the possibilistic product division,and proposes a novel interval type-2 unsupervised possibilistic fuzzy clustering algorithm driven by deep structure information.Firstly,the Fuzzy Possibilistic Product Partition C-means clustering algorithm integrates fuzzy membership and possibilistic typicality in the form of a product.Secondly,the Unsupervised Possibilistic clustering algorithm is combined with the clustering number in objective function to improve the effectiveness of algorithm in processing different types of data sets.Finally,the interval fuzzy uncertainty is updated by adaptive weighted distance constructed by master-slave neighborhood information.Through a large number of experimental tests,it is verified that the proposed algorithm has good segmentation performance in different types of noise images,has strong robustness,and has certain advantages in practical application scenarios. |
