Minute Feature Analysis in Speckled Imagery

This paper tackles the problem of estimating the parameters of relevant distributions that describe speckled imagery. Speckle noise appears in data obtained with coherent illumination, as is the case of sonar, laser, ultrasound-B and synthetic aperture radar images. This noise is non-Gaussian and non-additive and, therefore, classical techniques of processing and analysis may fail. A universal parametric statistical model has been proposed for such data, and numerical issues arise when estimating its parameters. In particular, the usual techniques for optimization and for solving systems of non-linear equations often fail to converge and/or to produce acceptable results, specially when dealing with small samples. An alternated method is proposed and assessed, and it is shown to produce sensible results. As an application, real and simulated data are analyzed. We show that the discrimination of minute features in synthetic aperture radar images can be performed using the proposed procedure.