DefinePK

Abstract

A novel method for unsupervised change detection in multi-temporal satellite images using Gaussian mixture model (GMM) with spatial information is proposed. This approach is based on three steps. Firstly, the difference image between two Synthetic Aperture Radar (SAR) images of the same area taken at two different times is obtained using the standard log–ratio operator. Secondly, a preprocessing step of anisotropic diffusion is applied to the difference image. Thirdly, Gaussian Mixture Model is used for segmentation of the difference image in which the parameters are estimated using Expectation algorithm. The standard GMM considers each pixel as independent and hence the segmentation is sensitive to speckle noise present in the SAR images. To incorporate the spatial information in segmentation, anisotropic preprocessing is done and also the posterior probability computed in the M step is weighted with the mean filter. The proposed method is tested on four sets of multi-temporal images. The obtained results demonstrate the effectiveness of the method in obtaining higher change detection accuracies compared to the related methods.