Efficient modelling of SAR texture with a gamma-inverse gamma distribution for MAP-based speckle suppression
The precise statistical modelling of synthetic aperture radar (SAR) texture is crucial while formulating maximum a posteriori (MAP) filter for speckle suppression. In this study, the authors introduce an SAR texture model considering the mixture of gamma and inverse gamma distribution as an approximation to a generalised inverse Gaussian (GIG) distribution, which suitably portrays areas with a varying degree of heterogeneity. An estimator is proposed for the model parameters using the expectation–maximisation (EM) algorithm. Cramer-Rao bounds are also derived for the model parameters to evaluate the effectiveness of the EM estimator. Furthermore, the model is experimentally established as an approximation to the GIG distribution through Monte Carlo simulation. Suitability and applicability of the model is then validated through 1-look real clutter and multilook synthetic clutter data over textured areas. Utilising the above model as prior density, the authors proposed an MAP filter for speckle suppression in SAR clutter data from areas of a diverse kind. Finally, the effectiveness of the -MAP filter is assessed using different statistical measures and is found superior over the MMSE-based Lee, Kuan filters and MAP-based -MAP, -MAP, -MAP, CE-MAP filters in terms of speckle suppression and mean preservation.