Change Detection In SAR Images Based On MRF Model And Fuzzy Clustering

Change detection in SAR images involves the analysis of two registered multitemporal remote sensing images acquired in the same geographical area at two different times to aim at identifying land cover changes that have occurred in the study area between the two times considered.SAR imaging has been developed rapidly and the resolution of SAR image has also been increased.Compared to optical remote sensing,SAR sensors are independent of atmospheric and sunlight conditions,which make the change detection in SAR images becoming very attractive.Now,SAR image change detection has been widely used in military and civilian fields.Due to the coherent imaging mechanism,the SAR images will inevitably suffer from the presence of speckle noise,which will bring error to change detection results.Therefore,how to suppress the speckle noise in the SAR images has been an urgent problem in SAR image change detection.The spatial correlation between pixels can be considered to suppress the speckle noise effectively.Markov random field(MRF)can carry on the spatial context information of the images,which greatly improved the accuracy of change detection.How to keep more detailed information and accurate positioning of changes in the detection results is also important.The whole image in MRF model is considered as stationary images.Therefore,the change detection in SAR images cause that the loss of detail information and the reduction of efficiency.FCM clustering methods are also used for image segmentation due to their easily applied advantages.However,the classic FCM clustering method is based on two assumptions.One is that the pixels are independent of one another and the other is the cluster prototypes are stationary.These two assumptions make FCM clustering very sensitive to the noise and poor details-preserving.On the basis of the analysis on MRF model and fuzzy clustering and the two difficulties in SAR image change detection discussed,this paper puts forward a SAR image change detection algorithm based on MRF model and FCM clustering.Firstly,the nonlinear diffusion filter is used to filter the difference image generated by multitemporal SAR image,and then the multiscale images are from the coarse scale to the fine scale is obtained.In order to use the the guidance between scales,we treat MRF model as fuzzy clustering and reconstruct the objective function.The suppression of speckle noise is enhanced by the integration of the membership of different scales.At the same time,cluster prototypes are treated as adaptive to strengthen the ability to detect detailed information in the clustering process.Such processing helps to enhance the details-preserving purpose in the method.Fusion strategy is adopted to reduce the over-segmentation caused by local adaptive features.The advantage of this algorithm is that it can reduce the speckle noise as well as improve the accuracy of detection.The experiments performed on the real SAR images demonstrate the effectiveness of the proposed method in both speckle noise reduction and details-preserving.