In the folklore, MEM is criticized for resolution that depends on the signal-to-noise ratio. But there are sound theoretical reasons to believe that this effect is common to all nonlinear algorithms that know about noise (Andrews & Hunt 1977). If you want to ``fix'' the resolution in MEM, the best answer is to do as is done in CLEAN: convolve the final MEM image with a Gaussian beam of appropriate width to smear out the fine scale structure, and add the residuals back in.
There are occasions when the super-resolution exhibited by MEM images is reliable, although it is not yet feasible to predict these in advance.
1996 November 4
10:52:31 EST