5400 Grain Dissolver and MTF impact?

[WD]

Hi,

Wondering if anyone out there has done an MTF measurement with the Minolta
5400 both with and without the 'grain dissolver' enabled?

The reason I am asking is that I have been playing around with a diffuser
plate with another scanner. I have seen some remarkable improvements
in some frames, but I have seen instances where it seems the result is a bit
softer (to the eye). Since the Minolta is somewhat of a controlled environment
(assuming same unit, same operator, same film) I was wondering if anyone
has attempted to measure the impact of the 'grain dissolover' on MTF?

Thanks

W

 

[Alan Browne]

Hi,

Wondering if anyone out there has done an MTF measurement with the Minolta
5400 both with and without the 'grain dissolver' enabled?

The reason I am asking is that I have been playing around with a diffuser
plate with another scanner. I have seen some remarkable improvements
in some frames, but I have seen instances where it seems the result is a bit
softer (to the eye). Since the Minolta is somewhat of a controlled environment
(assuming same unit, same operator, same film) I was wondering if anyone
has attempted to measure the impact of the 'grain dissolover' on MTF?

I can't give figures, but I agree that sections of GD'd images appear slightly
softer. Following re-sizing (25% by area or less) and USM for print, however,
this softening is not distinguishable in the final print. OTOH, once re-sized
(down) for printing, most grain is compressed out in any case. If you print 300
dpi of a full resolution scan, I'm sure you would see the grain or the softer
image from GD. Some films are of course better in scan than others.

Cheers,
Alan.

 

[Fernando]

Wondering if anyone out there has done an MTF measurement with the Minolta
5400 both with and without the 'grain dissolver' enabled?

I'm gonna do that, as soon as I figure where I buried my slanted edge
test slide!! :'(

My subjective impressions are that GD is of no help. Grain is softer
with it, but I see some edge softening, too. And scanning time
skyrockets.
But I plan to do the test, and will post it here.

Fernando

 

[Fernando]

Wondering if anyone out there has done an MTF measurement with the Minolta
5400 both with and without the 'grain dissolver' enabled?

OK, I resorted to a less-than-ideal target slide, so absolute MTF
values are to be disregarded; still, I think the test it's OK in order
to evaluate the difference between non-GD and GD scans.

Without GD, I have an MTF50 value of 53.8 cy/mm.
With GD, the value drops to 49.2 cy/mm.

So we're talking about a 8-10% difference.

(Imatest 1.2.1, scans made with Silverfast AI 6.2.3d3 trial, manually
focused)

Bye!

Fernando

 

[Kennedy McEwen]

Hi,

Wondering if anyone out there has done an MTF measurement with the Minolta
5400 both with and without the 'grain dissolver' enabled?

The reason I am asking is that I have been playing around with a diffuser
plate with another scanner. I have seen some remarkable improvements
in some frames, but I have seen instances where it seems the result is a bit
softer (to the eye). Since the Minolta is somewhat of a controlled environment
(assuming same unit, same operator, same film) I was wondering if anyone
has attempted to measure the impact of the 'grain dissolover' on MTF?

Well I'll start by saying that I don't have a Minolta scanner or the
Grain Dissolver to test, but the whole principle behind the grain
dissolver is that it reduces the MTF, particularly the MTF above the
Nyquist of the sampling frequency, ie above about 100cy/mm.

The grain dissolver is *NOT* intended to reduce or eliminate grain - it
is intended to reduce and/or eliminate grain aliasing. Aliasing occurs
because the MTF of the CCD and the optics are greater than zero above
the Nyquist frequency of the sampling system. In an ideal world, there
would be a nice convenient way of optically cutting the MTF of the
scanner to zero above Nyquist whilst maintaining a near 100% MTF below
Nyquist. Unfortunately this isn't an ideal world and the lens and pixel
themselves have MTFs which are not only less than 100% at spatial
frequencies below Nyquist, but are usually much greater than zero above
Nyquist. This results in aliasing.

In digital imaging, the main impact of aliasing only affects specific
types of subject - those containing regular patterns, where aliasing
causes lower spatial frequencies to be produced in exactly the same
manner as Moire discovered with gauze. In film scanning though, the
effect of aliasing is more general, exaggerating the size and contrast
of the grain in the image. The solution to aliasing is to achieve the
basic Nyquist criteria for sampling - no image detail reaches the
sampling system, in this case the pitch of the CCD and stepper motor,
which has a spatial frequency greater than half the sampling density
itself.

One simple way of achieving this is to defocus the lens in the scanner,
ensuring that the MTF above Nyquist is attenuated to a level where
aliasing is eliminated. This, however, has an impact on the spatial
frequencies that below Nyquist, causing the image to look soft. A
better method is a scattering medium in the light source, such as the
grain dissolver, since this has the effect changing the light from a
collimated to a diffuse source - which has a consequential effect on the
MTF of the lens. However, it is inevitable that, even when optimally
designed, some reduction in MTF in the useful region below Nyquist must
be sacrificed to attenuate MTF above it and reduce aliasing.

It still surprises me how many people think a single pixel standing out
from the background indicates a sharp image, when all it really
indicates is a grossly undersampled image where aliasing abounds. Have
you ever looked at the image of a single point on a uniform background
looks like when reproduced by a system where the MTF is 100% up to
Nyquist and 0% beyond it? Most people would consider that to be soft,
but it is actually the best that can physically be achieved when
aliasing is eliminated. Consequently, the objective is rarely the
elimination of aliasing, merely its control to acceptable levels.
Nevertheless, that *always* result in some MTF degradation.

 

[Kennedy McEwen]

OK, I resorted to a less-than-ideal target slide, so absolute MTF
values are to be disregarded; still, I think the test it's OK in order
to evaluate the difference between non-GD and GD scans.

Without GD, I have an MTF50 value of 53.8 cy/mm.
With GD, the value drops to 49.2 cy/mm.

So we're talking about a 8-10% difference.

That is rather misleading since the effect of the grain dissolver is to
change a near collimated light source into a near diffuse one and hence
the majority of the impact is well above 50cy/mm, which corresponds to
about half the Nyquist of the Minolta scanner. It would be interesting
to see what the effect of the grain dissolver was at 80% of Nyquist, or
on the MTF20 frequency - I expect that to be a lot more dramatic than a
mere 10%.

 

[Dierk Haasis]

So we're talking about a 8-10% difference.

Now, I know everyone goes through a "test phase" at one time of his
life, mine is some years behind me, luckily. It may also be the case
that certain photographic intentions are very close to limiting
factors as found in laboratory tests.

But, do these numbers really affect your images in a way noticeable
under normal conditions? My experience is that much more blurring
comes in on the photographer's end; then there is lenses, film
(material), physical limitations, handling, eye irregularities (short
sightedness etc.) ...

Reminds me that one does not necessarily need more resolution when
enlarging - all depends on the proper (and usual) distance when
looking at a picture. Which was known already in ancient time and
brilliantly used by Pointillists.

 

[Fernando]

That is rather misleading since the effect of the grain dissolver is to

I don't know what to say, I simply did the test the Original Poster
asked for.
These are the results I obtained.
I repeated the test twice, selected 2 different sets of Regions of
Interest (of course the same ROI was chosen for the GD run and the
non-GD run), even tried re-focusing on different regions of the test
target, and the results were consistent: between 8% and 10% of
difference.
Enough for me, but of course if others want to repeat the tests and
publish the results, I'll be happy.

Fernando

 

[Fernando]

But, do these numbers really affect your images in a way noticeable
under normal conditions?

Yes. I select my best shots to do very large prints.
I use hi-res slide film. My shots are often taken with tripod,
self-timer and mirror lock-up, I set the aperture to optimal values
for my lenses, and I scan at 5400dpi.
Every bit of details counts for me.
One day I'll maybe able to afford a better scanner, but for the
moment, I try to pull out every ounce of performance from my DSE5400.


Of course other may have different needs.

Fernando

 

[Robert Feinman]

of the sampling frequency, ie above about 100cy/mm.

The grain dissolver is *NOT* intended to reduce or eliminate grain - it
is intended to reduce and/or eliminate grain aliasing. Aliasing occurs
because the MTF of the CCD and the optics are greater than zero above
the Nyquist frequency of the sampling system. In an ideal world, there
would be a nice convenient way of optically cutting the MTF of the
scanner to zero above Nyquist whilst maintaining a near 100% MTF below
Nyquist. Unfortunately this isn't an ideal world and the lens and pixel
themselves have MTFs which are not only less than 100% at spatial
frequencies below Nyquist, but are usually much greater than zero above
Nyquist. This results in aliasing.

I think what you are saying is that effectively the GD changes the contrast
at certain levels of detail. This is similar to the USM which also effects
the contrast only with certain image details. By sending less collimated
light through the slide the contrast for individual dye clouds is reduced
and thus they become less noticeable. Large picture elements such a real
image detail are not as much affected. The net result is a perceptual
improvement. Whether this can be seen in the final output is, of course,
subject to many other factors.
Personally I haven't found the GD to do much for me, but I tend to shoot
film in the 200-400 ISO range and thus the dye clouds may be too large
to be modified.
It would be interesting to see some scans with different types of film
(positive and negative) with and without GD.

 

[Kennedy McEwen]

But, do these numbers really affect your images in a way noticeable
under normal conditions?

If that was *all* that the difference was, then your implied comment
would be true. As I suggested though, the 50% point of the MTF curve is
only relevant in comparing similarly shaped MTF curves and the whole
point of the grain dissolver is that it changes the shape of the MTF
curve, making this measurement meaningless.

 

[Kennedy McEwen]

of the sampling frequency, ie above about 100cy/mm.

I think what you are saying is that effectively the GD changes the contrast
at certain levels of detail.

Change that 3rd last word to "dimensions" and I agree. ;-)

This is similar to the USM which also effects
the contrast only with certain image details. By sending less collimated
light through the slide the contrast for individual dye clouds is reduced
and thus they become less noticeable. Large picture elements such a real
image detail are not as much affected. The net result is a perceptual
improvement. Whether this can be seen in the final output is, of course,
subject to many other factors.
Personally I haven't found the GD to do much for me, but I tend to shoot
film in the 200-400 ISO range and thus the dye clouds may be too large
to be modified.
It would be interesting to see some scans with different types of film
(positive and negative) with and without GD.

Certainly there is a limit to the granular dimensions that the change
from collimated to uncollimated light can affect - and this is
determined by the f/# of the lens in the scanner. With larger dye
clouds of fast film the effect will be much less significant - because
they are aliasing less in the first place, not because of their size but
the softness of their boundaries. This relates to my comment on the
MTF50 assessment that Fernando did. The spatial frequencies that the
MTF falls to 50%, as measured by Fernando, are around half Nyquist for
the Minolta, but aliasing occurs because of significant MTF above the
Nyquist frequency, 106cy/mm in the case of the Minolta. So the effect
of the grain dissolver would be expected to be small at half Nyquist,
corresponding to coarse, well resolved detail in the image (true grain)
while it would be expected to be very significant at high spatial
frequencies, corresponding to resolved but aliased detail (aliased
grain).

 

[Kennedy McEwen]

I don't know what to say, I simply did the test the Original Poster
asked for.

The point I am making is that you only did a small *part* of the test
the original poster asked for - you only measured (or at least provided)
the MTF at one spot amplitude, and that occurred at a spatial frequency
where the effect of the grain dissolver *should* be minimal, and
confirmed it to be so. MTF is a *curve*, a series of numbers, not a
single number.

What is the effect of the grain dissolver on MTF closer to Nyquist,
where the main advantage of the Minolta's higher resolution lies?

 

[WD]

Great responses from everyone thanks.
Regarding my personal impressions,
Sometimes I am convinced there is virtually no true softening, other
times
it seems there may be just a little bit.

Regarding the MTF effects, I appreciate the discussion as to the
diffuser
acting as in effect an optical anti-aliasing filter (in the linear
sampled
system model). However, are there also additional perhaps non-linear
things going on? For example the Callier effect I believe is not a
'linear
system' phenomenon. Or scattering off 'trapped air bubbles'
(peppergrain
effect) or scattering off scratches on the emulsion. I guess it comes
down to scattering vs. pure absorption/transmittance. Scattering
of collimated vs. non-collimated light sources creates a different
final result. Does this difference go beyond the linear system modelor
cannot in fact be truly modeled as a linear system filter?

Finally, alot of the above is academic for sure. What I have seen
doing
A/B comparisons between frames scanned with and without a diffuser, is
that some seem to have no difference, whereas others there is quite a
difference. The difference IMO is primarily one of defect reduction
(specs, scratches, grain (aliasing?) etc.) I have seen this in black
and white negs. color negs. and color positives.

W

 

[Fernando]

What is the effect of the grain dissolver on MTF closer to Nyquist,
where the main advantage of the Minolta's higher resolution lies?

No GD MTF at Nyquist = 0.155
GD MTF at Nyquist = 0.119

Fernando

 

[Kennedy McEwen]

No GD MTF at Nyquist = 0.155
GD MTF at Nyquist = 0.119

Which is more like a 30% drop in MTF, which *is* significant.

 

[Mendel Leisk]

of the sampling frequency, ie above about 100cy/mm.

I think what you are saying is that effectively the GD changes the contrast
at certain levels of detail. This is similar to the USM which also effects
the contrast only with certain image details. By sending less collimated
light through the slide the contrast for individual dye clouds is reduced
and thus they become less noticeable. Large picture elements such a real
image detail are not as much affected. The net result is a perceptual
improvement. Whether this can be seen in the final output is, of course,
subject to many other factors.
Personally I haven't found the GD to do much for me, but I tend to shoot
film in the 200-400 ISO range and thus the dye clouds may be too large
to be modified.
It would be interesting to see some scans with different types of film
(positive and negative) with and without GD.

Just got an Elite 5400, 2 or 3 weeks back. I've found the only way to
avoid "artifacts" at sharply defined light-to-dark edges, when
scanning Kodachromes with ICE enabled, was by using the Grain
Dissolver.

 

[Fernando]

Which is more like a 30% drop in MTF, which *is* significant.

I always said I saw a significant drop in apparent resolution (on
high-frequency details) when using GD.
In fact, I never use it.

Fernando

 

[Kennedy McEwen]

I always said I saw a significant drop in apparent resolution (on
high-frequency details) when using GD.
In fact, I never use it.

But is that apparent drop in resolution a direct consequence of
restricting the MTF to avoid aliasing? In other words, are you
expecting more real resolution that the sampling density can
realistically support? I suspect that most digital image users fall
directly into this trap. It is very educational to view an ideal point
source for a perfectly sampled and filtered image. I suspect most of
the readers of this forum would consider it to be "soft", which merely
reflects a general over-expectation for the sampling densities we are
using.

 

[Ole-Hjalmar Kristensen]

Yes. And that leads to a fundamental problem with film scanning. In
order to extract maximum information with no aliasing for a given
sampling rate, you have to have a very steep *optical* low pass filter
at the Nyquist frequency.
Or alternatively, a less steep optical filter, and sample with a
higher density, then do low pass filtering and resampling digitally.
You cannot expect today's 4000 dpi film scanners to deliver 4000 dpi's
worth of information without aliasing.

KM> But is that apparent drop in resolution a direct consequence of
KM> restricting the MTF to avoid aliasing? In other words, are you
KM> expecting more real resolution that the sampling density can
KM> realistically support? I suspect that most digital image users fall
KM> directly into this trap. It is very educational to view an ideal
KM> point source for a perfectly sampled and filtered image. I suspect
KM> most of the readers of this forum would consider it to be "soft",
KM> which merely reflects a general over-expectation for the sampling
KM> densities we are using.

KM> --
KM> Kennedy
KM> Yes, Socrates himself is particularly missed;
KM> A lovely little thinker, but a bugger when he's pissed.
KM> Python Philosophers (replace 'nospam' with 'kennedym' when replying)