Membership Correction Based FCM Algorithm And Its Application To Image Segmentation

Fuzzy clustering is an important unsupervised learning method,which is widely used in many fields of science and engineering.Among the fuzzy clustering algorithms,fuzzy C-means algorithm has become one of the hotspots of unsupervised learning because of its high efficiency and practicability.However,the algorithm sets normalized constraints to memberships,which make it more sensitive to noise and abnormal points,and degenerate the clustering performance of the algorithm.In order to solve this problem,this dissertation propose a novel fuzzy C-means clustering algorithm using membership correction from the perspective of relaxing the normalized constraints of memberships.First,the normalized constraint of the fuzzy C-means algorithm on the sum of memberships of a single sample is relaxed to the sum of membership degrees of all samples,and then the sensitivity of the algorithm to noise and isolated points is reduced.Then,a new amendment method of membership is proposed to solve the problem that the memberships are too different after the relaxation of the constraints.The proposed method can modify the membership to a reasonable range.That is,it can effectively avoid the issue that some samples belong to a cluster of their own because of too large membership,and it can also avoid the difficulty of selecting the termination threshold of iteration due to too small membership,and ensure that the constraints of the total membership of all samples are always satisfied in the iteration process of the algorithm.The numerical simulation results show that the proposed algorithm achieves more accurate clustering results on UCI(University of California-Irvine)data sets,compared with other clustering algorithms based on membership modification.Furthermore,since image processing is one of the most important applications of fuzzy clustering algorithm,this dissertation applies the proposed algorithm to image segmentation to verify the performance of the algorithm.First,In order to depict the image in detail,pixel gray value,neighborhood gray mean,local similarity and gradient are used as pixel attributes.It can describe the image information more comprehensively compared with the conventional method which only uses the gray value of the pixel.Then,four commonly used clustering validity indexes are used: partition coefficient,partition entropy,Xie-Beni index and DBI index to determine number of image clustering.Finally,the algorithm is applied to imagesegmentation.The segmentation results for multiple images show that the proposed method can achieve good segmentation results,and the segmented image has better pixel equalization.