Remote Sensing Image Segmentation Based On Information Clustering

The traditional clustering algorithm depends on clustering feature.When used to segment images its incomplete definition of clustering feature usually leads to reducing the accuracy of image segmentation and increasing noises sensitivity.Therefore,a remote sensing image segmentation algorithm based on information clustering is presented in this thesis,which abandons the concept of using clustering feature to express clustering and defines global similarity with mutual information.First of all,image feature model and image label field model is built respectively using Gaussian distribution and Potts model;Then,several(or two)pixels in a cluster are selected to form multi-tuple(or two-tuple)and all possible multi-tuples(or two-tuples)are employed as features to represent the cluster;And then mutual information of grayscales of pixels within each multi-tuple(or two-tuples)is calculated to characterize the similarities in the homogeneous region and dissimilarities between homogeneous regions;The objective function is formulated by balancing the above similarity and dissimilarity;By maximizing the objective function,remote sensing image segmentation can be obtained.Simulated and real remote sensing images are segmented by the proposed algorithm.The results show that the proposed algorithm can realize effectively the optimal segmentation of region,and the proposed algorithm has higher precision of image segmentation and better noise immunity.