All depends on what you want to achieve - pretty pictures, or useful and accurate measurement data.
The ultimate way of eliminating LP is moonless, dark site away from city lights.
However.. this is ideal situation, sometimes impossible to arrange
Stellar objects are very small, almost point like - so spreading their light across wider surface using objective longer focal length (as long as the size of the star image is comparable or smaller than the size of the pixel) will not significantly reduce surface brightness (number of photons per pixel) of the star-like object image.
So, to reach faint stars in LP affected areas, the best strategy is just a bigger telescope. If you compare two OTA's featuring the same F-number, the bigger one will show stars as brighter spots on a similarly LP-illuminated background. And, the additional benefit of this approach is the reduction of exposure times and higher resolution.
Nebulae, being spread across wider area, are affected by F number and aperture just like the LP.
But if their spectral contents are different from those of LP, for them to stand out, theoretically the right filter could be used to remove unwanted spectral content, introduced by LP.
I think there are two issues here:
- filtering affects reflective nebulae the same way it affects LP.
- Emission nebulae, being narrow-banded in their emissions, should survive filtering nicely (and they do), so aggressive narrow band (for example H-alpha) filtering definitely make (hydrogen emission nebulae it this example) stand out nicely, even in full Moonlight.
But, wide-band LP filtering for colour images must be different.. it is very complex in terms of spectral response [to remove most common LP spectrum, which in practice varies from place to place - some cities use high pressure sodium lamps, some other use mercury.. or combination of anything available on totally uncontrolled (in terms of LP) market], and there is also some attenuation present in pass band..
Now, with results I have seen - they were not good (or, better to say, they were not up to my expectations) probably because of the following reasons:
- removing the parts of the spectrum from reaching the sensor caused colour imbalance
- restoring the colour balance to the similar level by stretching the histogram in particular colour channels introduced some additional quantisation noise.
On the other hand, apparently (at least according to my experience) removing the background (LP) on images in DPP had similar effect as using LP filter+restoring colour balance, but at no $cost for filters.
Longer exposures and bigger number of them for stacking will help here to increase the S/N - so histogram stretching and removal of the background produces more eye-pleasing result.
And my standards for image appearance are probably much less demanding.. as long as I can extract some useful photometric data from my images, I am happy.