The interpretation and simulation of clutter textures in coherent images

The author presents an analysis of the properties of coherent images of clutter texture expressed in terms of correlated K-distributed noise. Examples from synthetic-aperture radar (SAR) and sonar are considered. In each case a full two-dimensional treatment is given in terms of theoretical results for the single-point intensity moments and the two-point autocorrelation function (ACF). The correlated surface is assumed to have a gamma-distributed noise term, with either Gaussian or Lorentzian spectrum, which may be mixed with a constant term at an offset frequency to give an oscillatory behaviour. This surface is then imaged through a Gaussian instrument function to yield a correlated K-distributed intensity. These analytic forms are compared with real SAR and sonar data. Having demonstrated the validity of the clutter model they next present a method for simulating such clutter textures based on a two-dimensional linear filter. This can be chosen to reproduce the full second-order statistics exactly. Higher-order statistics are no longer K distributed, however. Examples of simulated texture are presented which closely resemble the original data.