Regional Remote Sensing Image Segmentation Based On MRF Labeling Field And Characteristic Field

In order to describe the texture features in homogeneous regions,improve the accuracy of segmentation of remote sensing images,the Markov Random Field(MRF)model is used for establishing the label field and characteristic field.In this paper,a new segmentation method based on MRF label field and characteristic field of remote sensing images is proposed.Firstly,the image domain is divided into several Voronoi polygons by using Voronoi partition technique.In order to characterize the relationship between neighborhood polygons' label,the MRF model is used to define the label field,and it is assumed to obey the Gibbs distribution.It is assumed that the pixels are independent of each other,and the Gaussian Markov random field(GMRF)model is used to define the characteristic field model.Based on Bayesian theorem,an image segmentation model based on MRF label field and characteristic field is constructed.After the model is set up,the Expectation Maximization(EM)/Maximization of the Posterior Marginal(MPM)is designed to estimate the parameters of the segmentation model and get the optimal partition.In order to verify the accuracy and effectiveness of the proposed method,the synthetic texture images,real texture images and remote sensing images are segmented.The experimental results show that the proposed method can describe the texture features within region more accurately,improve the ability of the algorithm to resist noise,and achieve high classification accuracy.