Abstract

Synthetic Aperture Radar (SAR) images are used in environmental monitoring and other applications, but this kind of images are contaminated with speckle noise, which is the reason why they are very difficult to automatically analyze. The speckle noise is non additive and non Gaussian, therefore must be modeled with special statistical distributions. In the recent years, the family of the $\mathcal{G}^0$ distributions has been used with success because it allows the modeling of the extremely roughness areas of a SAR image. The $\mathcal{G}_I^0$ distribution (for intensity data) depends on three parameters, $\alpha$, $\gamma$ and $L$ which are the parameters of texture, scale and equivalent number of looks, respectively. Under this model, different regions are characterized by the $\alpha$ parameter, hence the importance of the accuracy in the parameter estimation for this distribution. In this paper the roughness parameter is estimated by minimizing the geodesic distance on the parameter $\alpha$. The Gaussian distribution function is used and the mean and variance are estimated from a random sample. The method is assessed by using Monte Carlo trials by comparing the mean square error from several samples of data from SAR synthetic images. The next step is to calculate an expression for the geodesic distance using the $\mathcal {G} _I ^ 0$ distribution and apply it in real SAR image classification.