Detecting upper outliers in small gamma samples: A comparison of techniques

A common assumption for Synthetic Aperture Radar (SAR) data, is that the intensity return from textureless areas follows a Gamma law with mean λ > 0 and L > 0 looks. Many image processing techniques need to estimate these parameters using small samples. Unfortunately, the presence of discrepant observations in SAR data occurs frequently, even when dealing with small samples. This is mostly caused by the presence of a small strong backscatterer as is the case of, for instance, a corner reflector. Processing techniques based on the estimation of the parameters of the sample distribution are highly influenced by the presence of such outlying observations. Therefore, it is of paramount importance to identify them before applying techniques based on parameter estimation. We discuss test statistics designed to detect the presence of an outlying observation in a gamma sample, and we propose a new test based on an empirical estimate of the underlying distribution.