This application is most useful for smoothing image data containing diffuse emission, where a wide mask at low data levels smooths the diffuse emission sufficiently, but a progressively narrower mask as the data value increases means point sources are not blurred out.
Note that if the program is run on data including background, then this background is effectively included as "signal" when judging the degree of smoothing required to achieve the target s/n. Hence quite large target values of s/n (e.g. >10) may be required to achieve significant smoothing. In the case of background subtracted data, smaller target values will suffice. It is best to experiment.