One point I was never 100% clear about is if Dmax refers to an absolute
density or simply the ratio of light to dark that can be resolved in a
scan.
In general use it is an absolute measurement, referring the maximum
density of the film to perfect transparency. However, in the context of
scanners with an adjustable exposure, the term is often misused to mean
"Density Range", since a higher Dmax can be achieved by increasing the
exposure, at the expense of increased Dmin.
Thus a Dmax of 4.0 would be able to scan the theoretical range of
densities found in a slide it does not guarentee that the dark areas
would be scanned properly.
The problem is actually the precision to which the dark areas can de
determined. For example, even if the film only supports a Dmax of 3.6,
it can produce a large number of discrete tones between, say, a density
of 3.6 and a density of 3.3. If you have a scanner with a nominal Dmax
of 3.6 this usually refers to the noise floor of the analogue circuits
or the quantisation noise of the ADC. As a minimum, a scanner with a
Dmax of 3.6 must have a 12-bit ADC, however that will only reproduce 2
tones between the densities of 3.6 and 3.3 - clearly less than the film
can reproduce. Thus the shadows are reproduced much poorer than the
film, with a limited number of tones. This limitation is made more
obvious because the perceptual response is not linear - you can see much
finer graduations in tone in shadows than you can in highlights - and
the image must be converted from linear to perceptual scaling (normally
by gamma and gamma compensation) for you to view it correctly. This
stretches out the lightness in the final image between these two linear
tones that the scanner measured at original densities of 3.3 and 3.6.
A 14-bit scanner can reproduce 8 levels between these two densities -
better, but still considerably less than the film. Similarly a 16-bit
scanner can reproduce 32 levels, betters still. However if the total
noise of the 16-bit scanner was dominated by quantisation noise, then
such a device would be considered to have a Dmax of 4.8 - which you
would think ought to be easily capable of reproducing the shadows on a
film with a Dmax of no more than 3.6, but only just achieves it.
This is one of the reasons I prefer scanning negatives as I know my
scanner can cope with the range of densities found.
But with negatives you simply shift the problem to the other end of the
scale, the dynamic range. Here you have a situation where the contrast
of the image must be stretched, because the density range present on the
negative is much less than the density range of the final image they
represent. Consequently you amplify all the noise, including the
analogue noise, quantisation noise *and* the granular noise of the film
itself, to produce the image. Granted, that granular noise is amplified
when you print from a negative conventionally, but the scanner noise
sources must be much lower for negatives than for slides, because of the
contrast stretch. Now, of course, the noise is reproduced throughout
the perceptual range rather than being accumulated in the shadow regions
as it is with the slide, but overall it is worse for an equivalent
scanner, because of the contrast stretch implicit in the negative
process.