Parametric algorithms make, sometimes strong, statistical assumptions about the data being processed. They typically assume a normal distribution of the spectral properties of pixels within a class. This assumption does not always fit the practical application (or real conditions). In instances where the assumptions hold, the parametric algorithm will produce good results. This is not the case when the assumptions do not hold, which can especially be expected in complex landscapes with classes of high variance (Hansen et al. 1996). Examples of parametric classifiers include parallelepiped (PP), minimum distance (MD) and maximum likelihood classifiers (MLC) (Tso and Mather, 2009). Despite its reliance on assumptions, MLC is one of the most widely used classification algorithms in the remote sensing community (Foerster et al., 2012; Hansen et al., 1996; Wang, 1990).