Abstract
Data clustering has been widely applied to numerous real-world problems such as natural resource management, urban planning, and satellite image analysis. Especially, fuzzy clustering with its ability of handling uncertainty has been developed for image segmentation or image analysis e.g. in health image analysis, satellite image classification. Normally, image segmentation algorithms like fuzzy clustering use spatial information along with the color information to improve the cluster quality. This paper introduces an approach, which exploits local spatial information between the pixel and its neighbors to compute the membership degree by using an interval type-2 fuzzy clustering algorithm, called IIT2-FCM. Besides, a Semi-supervising Interval Type-2 Fuzzy C-Means algorithm using spatial information, called SIIT2-FCM, is proposed to move the prototype of clusters to the expected centroids which are pre-defined on a basis of available samples. The proposed algorithms are applied to the problems of satellite image analysis consisting of land cover classification and change detection. Experimental results are reported for various datasets of the LandSat7 imagery at multi-temporal points and compared with the results produced by some existing algorithms and obtained from some survey data. The clustering results assessed with regard to some validity indexes demonstrate that the proposed algorithms form clusters of better quality and higher accuracy in problems of land cover classification and change detection.
•Propose spatial information-based interval Type-2 Fuzzy C-Means clustering.•Propose semi-supervised interval type-2 FCM clustering for satellite image analysis.•Apply the proposed algorithms to land cover classification of multispectral images.•Apply the proposed algorithms to change detection of multispectral images.