Unsupervised Change Detection In SAR Images Based On Triplet Markov Field

Change detection on synthetic aperture radar(SAR) images is a process that utilizes the SAR images acquired over the same geographical area at different times to identify the differences between the two acquisitions. Being insensitive to atmospheric and illumination conditions, the SAR system is widely used in many applications, such as agricultural survey, forest monitoring, environment monitoring, natural disaster evaluation, and urban change analysis. With the development of SAR imaging technology, change detection in SAR images becomes more and more important. Due to the fact that SAR images suffer from the presence of the speckle noise, they have been the difficulties and the focus in change detection based on SAR images how to suppress the speckle noise and accurately detect changes in the region.The traditional change detection algorithms are generally the first to obtain a difference image between the multi-temporal SAR images, then this difference image as observation data processed using the threshold method. Only considering the information of a single pixel, there exist large detection errors due to the presence of SAR image speckle noise. Markov Random Field(MRF) model is employed in change detection for the reason that it can make full use of spatial contextual information between adjacent pixels, greatly improving the accuracy of detection. The whole image in MRF model is considered as a large homogeneous region, which simplifies the calculation. It is improper for SAR images to give the same weight to different textual regions, because the property of the SAR images is omitted. The method based on Triplet Markov Field(TMF) model is proposed to overcome this defect by introducing a third filed u to describe different nonsationarities in the image. Regions with different nonstationarites can be processed differently according to the field U. The definition of U is the core of TMF model, and how to use the U-field more effectively to describe the non-stationary characteristics is also the focus of our study. In addition, the commonly used first-order neighborhood system is not a good description of the non-stationary characteristics for noisy SAR image. It is necessary to use a larger neighborhood in order to make better use of image spatial context information.An unsupervised change detection method base on Triplet Markov Field Model with ahigher order neighborhood in SAR image is proposed in this paper. The TMF model is powerful in the nonstationary synthetic aperture radar(SAR) image analysis, while a higher order neighborhood can maximize the impact of the elimination of speckle noise. The third ?eld U in the TMF model is rede?ned to describe homogeneous regions and heterogeneous regions. The corresponding prior energy of(X,U) is reconstructed to make full use of prior information more effectively. Experimental results on real SAR images validate the superiority of the proposed method.