All sci-hub would have to do in this case is download the same paper through three or more accounts (different institutions, networks, countries?) at three different times, rasterize them and keep the common denominator. If a pixel has no common denominator, they'd have to fall back to a default value. This is by no means a perfect method and it has its weaknesses and pitfalls, but the resulting PDF will be far less useful as a means to de-anonymize accounts using information from the PDF itself.
Publishers still have other sources of information to de-anonymize accounts if the multiple accounts/downloads aren't truly isolated from one another.
Quite trivially, actually, thanks to the good old analogue hole: https://news.ycombinator.com/item?id=30084193
All sci-hub would have to do in this case is download the same paper through three or more accounts (different institutions, networks, countries?) at three different times, rasterize them and keep the common denominator. If a pixel has no common denominator, they'd have to fall back to a default value. This is by no means a perfect method and it has its weaknesses and pitfalls, but the resulting PDF will be far less useful as a means to de-anonymize accounts using information from the PDF itself.
Publishers still have other sources of information to de-anonymize accounts if the multiple accounts/downloads aren't truly isolated from one another.