Research On Robust Credibilistic Fuzzy Clustering Algorithm

Image segmentation,as the basis of computer vision,is also an important prerequisite for image analysis and image understanding,but it is also a classic problem.Image segmentation classifies pixels into the different clusters so that pixels belonging to the same cluster have similar characteristics,such as grayscale,color,texture,brightness,and so on.Because of the uncertainty and fuzziness of pixel belonging attribution,the problem of image segmentation is extremely difficult.Therefore,many domestic and foreign image processing researchers have made great efforts in the field of image segmentation,and fuzzy clustering can describe the uncertain information of images well,so the algorithms that uses fuzzy clustering to solve the problem of image segmentation has developed rapidly.Fuzzy C-means clustering(FCM)algorithm has become the most widely used fuzzy clustering algorithm because of its simple operation and fast convergence speed.However,the disadvantage of the poor anti-noise ability of the FCM algorithm makes it difficult to deal with the problem of image segmentation when images are corrupted by noise,so many scholars improve the segmentation performance of FCM algorithm from various perspective,or introduce new constraint variables to describe the effect of the sample on determining the clustering centers,or enhance the anti-noise ability of the algorithm by introducing local and non-local information.Among them,the credibilistic fuzzy C-means clustering(CFCM)introduces credibility into the FCM algorithm to deal with the problem of outliers.However,when the algorithm is applied to the field of image segmentation directly,the characteristics of the image pixel itself are not taking into account,which makes the algorithm sensitive to noise.In this paper,aiming at the defects of the existing credibilistic fuzzy clustering algorithms,a series of improved credibilistic fuzzy clustering algorithms are proposed.1.Aiming at the defects of the noise sensitivity of credibilistic fuzzy clustering algorithm,first of all,the modified similar measurement method by using the neighborhood spatial information and grayscale intensity information will be embedded into the credibilistic fuzzy clustering algorithm.Meanwhile,the idea of intuitionstic fuzzy entropy is also applied to the clustering algorithm to improve the anti-noise robustness.Finally,a robust credibilistic intuitionistic fuzzy clustering algorithm is proposed to solve the defects of the credibilistic fuzzy clustering algorithm.2.Aiming at the problem that the anti-noise ability of the credibilistic intuitionistic fuzzy clustering algorithm is weak,consider introducing local information into the existing credibilistic fuzzy clustering algorithm to improve the anti-noise robustness and ensure the preservation of detail information.For the parameter selection problem existing in the modified similar measurement method,an adaptive weight factor is constructed by using the homogeneity of pixel intensity of the neighborhood window.Finally,a robust credibibilistic fuzzy local information clustering algorithm is proposed.3.Considering that pixels with low spatial distance have high characteristic similarity,so the information of mean-filtered image and the information of median-filtered image are introducing into the objective function of the robust credibilistic fuzzy local information clustering algorithm.Besides,in order to improve the anti-noise ability,by modifing the weight factor proposed by Gong,a more reasonable neighborhood weight factor is proposed.Finally,a credibilistic local information fuzzy clustering algorithm that integrates the characteristics of various local information is proposed,and makes full use of the spatial neighborhood information.Moreover,using kernel space distance measurement to replace the Euclidean space distance measurement,a credibiblistic fuzzy local information clustering algorithm using Hilbert kernel space is proposed.4.Because the possibilistic fuzzy clustering algorithm can handle the outliers better,consider integrating the idea of possibilistic fuzzy clustering into the credibilistic fuzzy clustering algorithm,and combine the advantages of both the credibilistic fuzzy clustering and the possibilitic fuzzy clustering to deal with the outliers,and the possibilistic and credibilistic fuzzy local information clustering algorithm is proposed.The algorithm uses the idea of multiplying the typical values by fuzzy membership to deal with the situation where images are corrupted by strong noise.In addition,the new local information weight factor is constructed by using the maximum grayscale,mean grayscale and median grayscale of pixels in the neighborhood window.The weight factor can deal with the situation well when images are corrupted by Gaussian noise and Salt & Pepper noise.