Research On Change Detection Of Remote Sensing Images Based On Mixture Of Factor Analyzers And Markov Random Field

Change detection of remote sensing(RS)images is to extract change information from two original RS images,acquired in the same geographical area at different times.With the rapid development of the RS technology,RS images contain rich spatial,texture,and geometric structure information.How to use adequately the interesting information and acquire effectively yet quickly the change information is the key point for change detection of RS images.This thesis mainly focuses on effective generation of change mask(CM)in an unsupervised way,and two change detection methods are proposed,which are elaborated as follows:(1)We first investigate the feasibility of factor analysis for change detection,and then extract difference features by the factor analysis,whose parameters are estimated by using the expectation-maximization(EM)algorithm.Besides,there is an overlapping area between the changed and unchanged classes,so an effective two-level fuzzy clustering method is designed to cluster the difference features.Subsequently,a binary CM is yielded.Finally,experiments on four groups of real RS images show that the proposed method can achieve better detection performance.(2)Based on Bayesian theory,we discuss the feasibility for change detection using mixture of factor analyzers(MFA)model.Specifically,the proposed method firstly uses the MFA to model the data distribution of difference image,where the parameters of the model are also estimated by using the EM algorithm.Then the binary CM is obtained by minimizing mean square error(MSE)criterion.Besides,considering that there is the local similarity in RS images,the detection performance could be further improved by incorporating the spatial information.Therefore,this thesis proposes a new change detection algorithm based on MFA and Markov random field(MRF).Finally,experiments on four pairs of real RS images verify the effectiveness of the proposed method.