Markov-Potts Prior Model and Fuzzy Membership Based Nonparametric SAR Image Change Detection

In order to improve the accuracy of synthetic aperture radar (SAR) image change detection, a novel unsupervised non-parametric method for change detection is described. The method treats the prior data and the observed data as two independent events and adopts a simple algorithm to realize and validate the effectiveness of the proposed method. Firstly, the prior distribution is obtained by MRF with Potts model of the initial classification result by k-means with the prior data. Secondly, the fuzzy probability is obtained through fusing gray value and texture feature fuzzy membership of the observed data. Meanwhile, the fuzzy probability is regarded as the data likelihood probability. Finally, by using the Bayesian formula and the independent distribution criteria to calculate the maximum a posteriori (MAP) probability, change detection can be regarded as the product of two probabilities of two independent events. Simulation results show that the proposed method effectively combines the gray and texture information of difference image, overcomes the shortcomings of using probability statistic model and parameter estimation, reduces the influence of speckle noise of SAR image and improves the accuracy of image change detection.