Theoretically, it would basically be just a matter of subtracting the "general" light background, and then apply a deconvolution to the remaining "signal" to reduce blurry areas of light into point-like stars.
However, deconvolution is one of those things that's great in theory, but almost impossible to pull off in practice, except under extremely controlled circumstances. And in this case, considering just how much dust/smog/etc. is in the atmosphere, and how it is non-uniform, I expect it would be simply impossible.
There's a lot of that kind of stuff that goes into signal-cleaning in observatories, especially those located near major cities (like the Lick Observatory outside San Jose). But it tends to be pretty site/telescope-specific, full of in-house custom code, so I'm not sure if there's anything good that a regular photographer could use to produce the same effect.
However, deconvolution is one of those things that's great in theory, but almost impossible to pull off in practice, except under extremely controlled circumstances. And in this case, considering just how much dust/smog/etc. is in the atmosphere, and how it is non-uniform, I expect it would be simply impossible.