BTW: choosing a single level between black/white doesn't sound like a good solution. AFAIK there are better ways available nowadays that use a dynamic threshold to determine which parts are black / white.
That's not what I suggested. You're thinking of two-tone, ie fax mode, which is evil even for simple text. I meant, choose the upper and lower scan cutoff levels to give true white and black in areas that are supposed to be white and black. I say 'supposed to be' because on paper they never actually are, unless you're printing with vantablack and surface-of-the-Sun plasma. But the publisher's intent was pure white and black, so it's valid to assign ffffff and 000000 codes to those pixels.
There still need to be gray levels between. Just
how many levels, depends on the context. For black and white text, where all that's needed is to preserve visually clean curves on character edges, 16 levels (4 bits/pixel) total is adequate with sensible pixel sizing relative to the font. For B&W photos, at least 256 and preferably 64K levels to avoid visible posterization effects. For full colour, then 24 bit or better.
But the main point is to remove visually insignificant noise in flat color areas, so PNG's RLL compression scheme can work best.
Btw, 'dynamic threshold' can't work for multi-page documents. It will adapt differently on pages of different content, resulting in digital page representations that look different when they should be the same. You have to do trial scans of representative pages, then choose a scanning and post-processing profile that works best for all of them, then stick with that one profile through all the work. Unless there are radically different types of pages, in which case you need a profile for each type.