A SAR Image Segmentation Method with Automatic EM and Nonlinear Anisotropic Diffusion

SAR image segmentation is an important preprocess in ATR,which influences the recognition effect of SAR ATR system.A SAR image segmentation method with automatic EM based on anisotropic smoothing is presented.Firstly,the original SAR image is smoothed using an anisotropic diffusion.Then a Gaussian mixture model is used to combine intensity information with spatial structure information for classification.Finally the class conditional probabilities are smoothed using an anisotropic diffusion to remove the influence of speckles.Simulating result shows that the approach is robust and automated.