5.4 Non-Gaussian speckle models

Non-Gaussian speckle is found when the Goodman model conditions are no longer met at every pixel scale, or if they are still met, when the surface is no longer stationary. In the former case, this is essentially due to a small number of scatterers (or the predominance of a few scatterers) and low roughness, as may be seen, respectively, on images of urban sites captured by sensors with very high spatial resolution and in the nearly specular reflection produced, regardless of the spatial resolution, by smooth surfaces normal to the line of sight. When the surfaces, while still satisfying the Goodman criteria, are no longer broadly stationary, the observed phenomenon is often referred to as "texture". We will also call such surfaces "heterogenous". Within the framework of stationary heterogenous surfaces of second order, the scalar product model may be used to model the properties of such surfaces in statistical terms. These models may also be used to reduce speckle, describe target detection clutter and for classification purposes.

Finally, we can also model the data without any consideration of the multiplicative model. This makes it possible to resolve detection, segmentation and classification problems based on the initial data, without having to resort to reflectivity R as an intermediary. In this case, the radar people would rather use the general term of clutter to characterize the response of the surfaces.