Image Segmentation Based On FCM Algorithm

Fuzzy clustering is widely used in image segmentation, for the method dowell for the uncertainty sample points. In this paper, we do some research forfuzzy c-means algorithm which is the most classic method in fuzzy clustering.Besides, we improve and optimize the algorithm, and the validity of the methodthat this paper proposed is verified by experiments.First, in order to perform the standard FCM algorithm, determining thenumber of clusters is necessary, but the method of this algorithm is subjectivelygiven by experience. Some scholars determine the number of clusters by usingthe number of gray histogram peaks, but due to the noise, each pixel belonging toa gray value exactly is indeterminate, which may cause errors. Fuzzy theory candescribe the degree that one pixel belongs to a gray value, so this paper proposesfuzzy histogram for gray images and fuzzy Histon histogram for color images.Second, for gray images, the noise resistant of the standard FCM algorithmisn’t good, because it’s only using gray level information. Combining with grayand spatial information, Chen has modified the objective function. Chen’salgorithm considers that the contribution that each pixel acts on the center pixelin the space is the same, but actually it’s different. According to this problem,this paper proposes a modified SFCM algorithm, making classification moreeffective.Third, the definition of the standard FCM algorithm about distance isEuclidean distance, which do well for the spherical structure of samples but haspoor effect for the non-spherical structure of samples. So we consider using theidea of kernel function with spatial information in this paper, in order to increasethe separability of samples through mapping samples into a high dimension.Finally, MSFCM algorithm in the paper will be applied to color imagesbased on Lab color space, and we analyze the role of parameters in the MSFCMalgorithm as it affects the quality of Lab color image segmentation.