Create a good StarMask, and add it (with PixelMath) to the first one or two Wavelet layers.
Or, start with a Lu channel extraction, and subtract the R layer from an appropriate Wavelet selection - leaving 'small scale' only (the 'R' wavelet layer having been adjusted to include only the largest 'nebulosity')
If you know anything about the Zone Processing System that Ron Wodaski recommends, then you are trying to confine different types of processing to different areas of your image (in a 2 x 2 'matrix' that has 'noise' on one axis and structure size' on the other).
So, at 'bottom left' you would have the dimmest AND smallest data. At 'top right' you would have brightest and largest data.
It seems to me that what you are trying to process is that section of the matrix representing ['biggest' but not quite 'noise free']. Were you interested in Stars alone, these would fall into the ['almost smallest' and 'noise free'] category, and the backgound itself would be ['smallest' and 'noisiest']
(and, in all of the above, substitute 'SNR' for 'noise' if you feel inclined to do so, but, in any case, read Ron's excellent book - he explains it FAR better than I can
)
Cheers,
