SAR image segmentation using wavelets and Gaussian mixture model

Synthetic Aperture Radar (SAR) segmentation is often acknowledged as a difficult task due to the presence of speckle noise because of which traditional segmentation algorithm fail to give satisfactory results. In this paper, Gaussian Mixture Model (GMM) along with the combination of wavelets is proposed for noisy image segmentation. First, texture feature are abstracted in the wavelet domain and according to the features of its distribution, it is filtered. Finally, the SAR image is segmented using GMM, the parameters of which are estimated by EM algorithm. The pixels are classified into different classes according to their probability belonging to each Gaussian distribution.