Contents I. Introduction
Fuzzy C-means (FCM) clustering method is intuitive and easy to implement in image segmentation, but it also has some shortcomings, such as large computation, algorithm performance depends on the initial clustering center, and the number of clusters must be determined, which limits its use in image segmentation. For this reference [1], it is proposed to use gray histogram statistical information, which greatly reduces the computation, but for color clustering of color images, multi-dimensional expansion of histogram computation costs a lot. Document [2] initializes the clustering center by hierarchical subtraction clustering to improve the calculation speed of color image segmentation, but does not further improve the image segmentation effect. Document [3] combines multi-resolution analysis with FCM to realize fast image segmentation at multi-resolution, but there is no solution to give the optimal number of clusters. In addition, in view of the fact that FCM algorithm is not suitable for clusters with non-convex shapes and is sensitive to isolated points, Mercer kernel is introduced into FCM algorithm to map input spatial data to high-dimensional feature space, thus better clustering is carried out in feature space [4].
