High Resolution Remote Sensing Image Segmentation Based On Interval Type-2 Fuzzy Theory

Image segmentation is a basic and important task for remote sensing image processing(such as feature extraction,object recognition,image classification).High resolution remote sensing image has more abundant surface detail information,which provides the basis for the accurate object segmentation.However,the uncertainty of pixel membership and that of segmentation decision are increased caused by more precise spatial scale,complex and diverse objects,and the lack of real surface coverage information,which bring new problems for high resolution remote sensing image segmentation,and not improve the image segmentation accuracy by using traditional image segmentation methods.To this end,the uncertainty of pixel membership,the uncertainty of segmentation decision and spectral measure distribution model of homogeneous regions are considered as research goals,this paper proposes a scientific problem of image modeling and segmentation based on type-2 fuzzy membership theory to research on the theory and practices to segment high resolution remote sensing image accurately and reliably.The main research contents are as follows: First,the characteristics of high resolution remote sensing image are analyzed to establish the gaussian model is considered as type-1 fuzzy model between the gray measurement of different regions and the membership target of the uncertainty relationship by supervised sampling to define the uncertainty of pixel membership.Second,the different ways of type-2 fuzzy model of homogeneous regions are built based on the membership function model of homogeneous regions to strengthen the uncertainty of pixel membership and that of segmentation decision;and the effects of segmentation decision caused by different models and different uncertainty region range are analyzed.They are reliable basis for segmentation decision.Third,type-1 fuzzy model and the upper and lower functions of interval type-2 fuzzy model are combined to build fuzzy decision model with the spatial relations to make segmentation decision.Fourth,the feasibility and effectiveness of the proposed method are verified by testing a variety of high resolution remote sensing images.Synthetic and real high remote sensing images are tested by the proposed method,Gaussian membership function,Maximum likelihood algorithm,FCM algorithm,HMRF-FCM algorithm and the proposed method in this paper.The qualitative and quantitative experiment results show the feasibility and superiority of the proposed approach.The proposed image modeling approach based on interval type-2 fuzzy theory effectively solve the segmentation problems caused by the uncertainty of membership of pixels and that of the segmentation decision to improve the segmentation accuracy,and provide an effective way for the interpretation of high resolution remote sensing images accurately.