A novel mathematical approach for the problem of CFAR clutter model approximation

Automatic Target Recognition (ATR) in Synthetic Aperture Radar (SAR) is a set of techniques which are able to detect, discriminate and classify objects present in the observed scene. Unfortunately the presence of speckle degrades target detection dramatically therefore denoising algorithms are necessary. Moreover sometimes other operations, such as incoherent averaging for instance, are used to increase the Constant False Alarm (CFAR) performance. Any operation as a consequence changes the background clutter model and often it is not possible to describe it in a closed and manageable mathematical form, therefore a direct solution of the Neyman-Pearson problem is not feasible and a suboptimal criterion, such as Exponential or Gamma-distribution clutter model etc., is usually adopted. A consequence of the suboptimal global clutter model choice is the reduction of the information content of the SAR images which can affect classifier performance heavily. This paper hence is concerned with a novel mathematical approach for a local approximation of the filtered SAR image Cumulative Density Function (CDF) in order to preserve/maximize the information content carried by a SAR/ATR system.