Optimum CFAR detection against compound Gaussian clutter with partially correlated texture

The paper deals with CFAR detection in compound Gaussian clutter with a partially correlated texture component. A theoretical high performance upgrade has been demonstrated using the ideal exact knowledge of this component in the CFAR scheme (`ideal CFAR´) in a paper by Watts (1985) for K-distribution. For practical application the authors derive some optimum local texture estimators, based on the closest range cells: and use the estimated values to set the detection threshold. The schemes differ for operating over the intensity or the logarithm and for using or not prior information about the texture correlation. In particular, a maximum a posteriori estimator is derived, which outperforms the usual cell averaging CFAR and provides always performance close to the `ideal CFAR´. The derivations are valid for all compound Gaussian clutters. The performances obtained with K and compound weibull distribution are compared, assessing the robustness of the proposed detection scheme