I. Introduction
For many Earth observation applications, remotely sensed spectral imagery is only useful after the derivation of surface reflectance from a given airborne or spaceborne instrument's radiance measurements. The atmospheric correction involved in this derivation is delicate and greatly affects the accuracy of the resulting spectral reflectance data [4]. Often, additional postprocessing techniques such as spectral smoothing, i.e., spectral polishing [5], or spatial averaging are used (if possible due to target homogeneity) to improve spectral accuracy at the cost of spatial resolution.