David J. Littleboy said:
Well, the published MTF curves certainly _predict the response incorrectly_
for low contrast targets.
No they don't - they present the MTF and that is independent of the
signal contrast. How you choose to interpret, or mis-interpret, the NTF
curve for your own use is your problem. Learn how to use the tool
before criticising it, I say!
The film is the measurement medium, and it's not responding. You can't
idealize away the noise: it's an inherent property of the medium. If a
section of your photograph has detail with 1.6:1 contrast _as projected on
the film_, then you won't see that detail.
Exactly, and I am not disputing that, although I do dispute your
suggestion that you are only concerned with source contrast in that
region. Your own results show final contrasts outside of that region
so, allowing for the MTF losses already identified, the source contrast
must have been significantly greater than you suggest.
Again, the complaint is that for the images I'm taking, I need low-contrast
performance. The theoretical MTF in response to a contrast 500 times more
than what I've got doesn't tell me what the film does in my images. The
signal that I get off the film has noise in it, and I don't have the option
of ignoring that noise.
And I am not suggesting anything different - every condition sets
limits, but understanding what those limits are and what secondary
effects they impose is just as significant. If, as you suggest (even
though I dispute it), that all of your source material is 1.6:1 contrast
and less then granular is, and always will be, your biggest enemy - and
that enemy will alias *very* significantly with both 2700ppi and 4000ppi
scanners.
I started by conceding you the image content, whilst pointing out that
the grain was also important - you seem to be intent in arguing that the
image is actually worth a lot less than my starting position on this.
;-)
??? I thought that the MTF was the contrast of the thing being measured. If
I'm measuring the medium, that's what I want to measure.
No - the MTF is the *relative* contrast between input and output (ie.
the *ratio* of output contrast to input contrast) of the medium, or
system comprising several media. MTF is *NOT* contrast, hence my
reference to the difference between the two in the previous quote! The
medium itself has no contrast - whether you wish to measure it or not.
Most of the signals we record on slide films in actual photography are areas
that fall in a fairly narrow range of zones. So the response that we
actually use and need to render our subjects (not just the edges) is the low
contrast response.
So why, if we are only interested in low contrast response, reduced even
further by lens MTF and film MTF, do we need scanners with 16-bit ADCs?
The contrast of the now obsolete 10-bit films scanner is 1000:1, and
with MTFs of camera lens, film, scanner lens, and scanner CCD summing up
to less than 5% for a significant part of the spatial frequency range,
that obsolete capability became saturated with 20,000:1 scenes!
Talking about the performance of a system abstracted away from the noise
seems very much the wrong thing.
Not at all - MTF is useful to determine the signal. The noise can be
determined from other parameters. It is quite common in sensor system
modelling to compute signal and noise completely independently and
combing them to create a signal to noise ratio - in fact I know of no
reputable analysis which does otherwise!
I doubt it. The shadows are lit by full open sky. That's a lot of light.
Shadows lit by open sky can only exist on the top of the units - the
shadows in your image have a significant range of levels. The higher
contrast in the digital image suggests that some of these are well below
1/1000th of the white levels after monitor gamma is taken account of.
Bart discussed this here before: it's really quite hard to create a 1000:1
contrast.
As I previously pointed out, if it was so difficult then we certainly
would not need the equipment, either film or scanner, that we do.
This argument is problematic because I don't really mean to say never: I
mean to say rare. Since low contrast is far more common, it's the low
contast performance of the film that I need to model.
I disagree - you need a film that copes with the extremes that you are
likely to throw at it. Any old muck can cope with the mundane - its the
stuff that copes better with the exceptions that gets the reputation for
performing.
Ah, I see the reason for the silliness: you've misunderstood what I'm trying
to say.
I don't think so - I see you repeating it several times in your
response!
Detail selected being low contrast, but still higher than your nominal
reference. The foreground detail, like all short range material, has
much less atmospheric attenuation and this much higher contrast.
Just in case you weren't aware, the atmospheric transmission attenuates
the highlights and elevates the shadows by scattering the available
illumination, such as ambient daylight.
For reference - two adjacent pixels in the foreground read RGB values of
189,195, 144 and 21,35,0. Combining these RGB values with a relativel
crude 1,2,1 weighting gives a gamma compensated contrast in excess of
8:1, and a linear contrast of over 40:1. Given the MTF of the system
components used, a source contrast approaching 1000:1 is not too far off
being present there in the example image that you chose to offer!
If I want to resolve the texture in the trees in a distant forrest, then the
contrast isn't a whole lot over 1.6:1. Even the rice in the foreground is
fairly low contrast.
Actually, from the figures presented above, it is close to and probably
even exceeds the figures that you claim never to encounter!
If one does the standard landscape on Velvia with
polarizer bit (that shot is Provia without), one doesn't get the contrast
from the specular reflections, and is left dealing with much lower
contrasts. It's the 1.6:1 performance that imposes the practical limits on
what I can get out of my film.
No it isn't - it is the MTF applied to all of the contrast present in
your image, not just those you care to assess.
I've tested all my Mamiya and Fuji 645 lenses, and even for my relatively
high contrast test scenes, I don't see any differences (observing with a 60x
microscope) from one stop down from wide open to f/16. As I've said before,
I can maybe persuade myself that f/22 is a tad softer.
Converse to conventional wisdom, large format lenses tend to be lower
performance than small format lenses. This happens for two key reasons.
Firstly, because they can get away with it, given the lower levels of
magnification of the film. Secondly because it is much harder to
maintain performance across the field, and it is generally preferable to
have a slightly inferior cross field performance than a cracking on axis
and poor edge performance. Your Mamiya lenses may well have sweet spots
down in the f/16 region - never getting any better with smaller f/#s -
however you will find 35mm optics considerably better, albeit over a
smaller area, and digital camera optics often even better still over an
even smaller area.
My 60x microscope doesn't show that. With Provia 100F for real-life images.
As mentioned earlier, you are looking at lens trade-offs which take
advantage of the larger format. Diffraction limit f/16 is only 10% at
100cy/mm, but f/11 is 33% and f/8 is 50%. If you see no difference then
clearly your lenses may be good as mid format optics, but they are
relatively poor in terms of 35mm and next to useless for small format
(3um pixel) digital media - where the small formats demand higher
performance, and would never have become acceptable without it.
Another issue is how does MTF change as one moves away from the plane of
focus.
Simple to compute - multiply the MTF with the FT of the circle of
confusion. That's the way MTFs work.
I suspect that what's going on here is that what we use for our
practical photography really isn't much over 30 lp/mm most of the time. Off
axis, normal to wide angle lenses crap out pretty badly even at 30 lp/mm and
f/8, and we actually use the off axis areas. And regions away from the plane
of perfect focus are obviously less contrasty as well.
A 30lp/mm upper limit would be abysmal on a 35mm format - even a flatbed
scanner would be able to discern the limitation of 1500ppi, so again yo
do tend to be considering the effect of mid format again, where the
larger sizes involved mean that depth of field is much more restricted
that the equivalent field of view small format lens. I have an 18mm
f/3.5 that is pretty extreme FoV on 35mm and certainly produces much
more than 30cy/mm on film. In fact, it is probably one of the sharpest
lenses I own at the upper and lower frame edge and although side edge
and corners do have noticeable fall-off, even they have significant MTF
at 30cy/mm.
OK. Fine. I get the content bit and you get the grain bit<g>.
It sounds, though, that including an antialiasing filter on the scanner CCD
would be a good idea.
But I bet most scanner users would scream bloody murder if mfrs started
doing that, though<g>.
I don't hear many Minolta 5400 owners screaming bloody murder about the
grain dissolver, which is just that, if a less than perfect example. I
accept that many flatbed users do scream about poor performance of their
units which incorporate anti-alias hyperCCDs, but I have yet to be
convinced that is not just cheap optics and a lack of understnading on
the user's part of what a properly sampled image looks like under
magnification. The whole point is that you don't see individual pixel
sized objects, but that is what the masses seem to expect, so they
complain that their scanners are poorly focussed.
Not total system: lens + film<g>. One problem with my counting pixels in
edges _in the scan_ is that that includes the scanner MTF.
It is the total system MTF that determines whether the image information
will alias, not just the lens + film.
The high-frequency components in the Fourier transform (of an edge) are
lower magnitude than the fundamental, they're attenuated by the lens, and
they're attenuated by the film. That's getting to be a lot of attenuation.
Their original amplitude is unimportant in terms of how they affect the
edge being smeared out, that comes down to how well they are reproduced
by the system.
Yes. But they are so grossly attenuated that they differ insignificantly
from the case were they not there.
Sadly that isn't always the case. I wish you were right, but it just
doesn't happen that much.
Again, this is my basic rant: worrying about image detail at sky high (note
the informal term<g>) lp/mm levels is barking up the wrong tree.
As stated right at the start of our debate, for image detail that is
often, but not always, the case - however image data isn't the full
story. Grain, drives you to higher resolutions to avoid aliasing issues
with very high source and final contrasts and, quite frankly, unless you
control grain at source before the image even reaches the sampling
sensor, that drives you to submicron sampling to ensure that all of the
potential aliasing signal is completely attenuated by the optical MTF.
Sure. But alias so that they are significant? I've not seen it. Have you
been able to demonstrate image aliasing in the Minolta scanner?
It is no secret that I don't have a Minolta scanner. ;-)
However I have produced aliasing on the Nikon LS-4000 in a couple of
real images. As I said some time ago though, it was the reduction in
grain aliasing that persuaded me to move to the higher 4000ppi
resolution from earlier Nikon scanners, rather than a marginal increase
in image resolution.
Note that the "random dots" I'm talking about are dots _IN THE SCANNED IMAGE
NOT THE FILM_.
In which case their fundamentals are not out of band, they are an
in-band alias of the originally out of band fundamental of the random
dots ON THE FILM! Nevertheless, depending on how you display the
scanned image, those final dots will have a certain amount of out of
band information too - otherwise they would not appear as dots, but
fuzzy splodges. ;-)