Scanner bake-off: initial wave of results compiled and published...

[Bart van der Wolf]

SNIP

Somehow I have the feeling that you count de difference in resolution

twice.

You are correct. Thanks for spotting that.

Your DSE-5400 gets an MTF20 value of 0.1963. This results in
0.1963 * 2 * 4844 = 1902 line/ph.
The LS-8000 gets 0.2762 * 2 * 3616 = 1997.

The "per Picture Height" measurement is a bit tricky with scanners, as
cropping may be different (which it is), although these two seem rather
close.

No matter what you do, you won't get more that 1902 lines (at MTF20) whereas
the LS-8000 gets 1997 lines. Scaling doesn't introduce new lines, and
should not drop lines. Unless, you are saying that higher scaling factors
would somehow result in a serious loss of sharpness.

No, you are right. Assuming the cropping is equal, in the 5400 case there
are 4844 scanned lines per slide's height, whereas in the LS-8000 case there
are 3616. So reaching the same ppi in output requires 34% more magnification
of the LS-8000 scan. That will reduce the LS-8000 from 0.2762 cy/px to
0.2062 cy/px, versus the 0.1963 of the DSE-5400, or 4.8% lower but with
lower contrast and less clipping.

Both scanners seem IMO a bit limited by the film. It would have been nice to
also have a very high drum scan result. That would reveal if there is more
that can be extracted.

Bart

 

[Bart van der Wolf]

What is the difference between the calculation you used and
the calculation in this test? I assume you followed the ISO standard.

No, my tests of scanners were conducted over a period of several years, and
they started out as optical microscope comparisons of lens+film, later
adding scanners to the imaging chain. The ISO standards for film scanning
are yet to be finalized
(http://www.iso.ch/iso/en/CatalogueDetailPage.CatalogueDetail?CSNUMBER=29702
&ICS1=37&ICS2=40&ICS3=99 for Reflective media has meanwhile become
available). I tested for absolute limiting resolution, zero contrast
difference. My reasoning being, if there is visible detail, then there is
visual resolution (and sharpening can enhance it).

The ISO 16067 standard, parts 1 (Scanners for reflective media) and 2 (Film
scanners) sets out to determine the SFR (Spatial Frequency Response) of the
scanner itself. The method is the same as used for ISO 12233 (Resolution
measurements for electronic still pictorial cameras,
http://www.iso.ch/iso/en/CatalogueDetailPage.CatalogueDetail?CSNUMBER=33715&ICS1=37&ICS2=40&ICS3=10,
yours for 116 Swiss Francs, US$ 93.19, EUR 75.93). However, the algorithm
used for calculating the SFR/MTF can also be used for the entire imaging
chain. It doesn't only determine limiting resolution (the ISO hints at 10%
modulation, I used 0%), but it calculates response for all frequencies up to
4x Nyquist (2x sampling frequency). That does require measurements of
scanner/digicam OECF (contrast, and linearizing it), and it also describes
the target to be used for automated evaluation.

But I can't find any on-line resources that describe that basic idea

behind the ISO standard.

Unfortunately the PDF work documents that used to be available on the PIMA
website, are copyrighted, and no longer available for us mere mortals on the
new www.I3A.org site.

Basically it (there are three standards that use the same algorithm on
device specific targets) describes the procedure for contrast normalized,
Spatial Frequency Function determination. It is based on sampling a slanted
edge and producing an MTF like curve that describes the Scanner's/Digicam's
modulation at a range of spatial sampling frequencies.

In your test, your scanner scores quite a bit higher than the LS-4000.
Did you somehow take the extra resolution into account?

No, just lp/mm on sensor at absolute limiting visual resolution
(http://www.xs4all.nl/~bvdwolf/main/foto/scan/se5400/se5400.htm example 1,
yes I have to update those pages but VueScan gives wrong Dmax). That would
still require normalizing for equal output size.

I have since developed improved targets that allow to evaluate cycles/mm at
a variety of angles (not only Horizontal&Vertical) versus the sensor array,
and thus more suited for direct sensor imaging (i.e. digicams). That
requires re-shooting the targets on current film generations (thus
potentially harder to compare to, but not invalidating, earlier results).

Bart

 

[Bart van der Wolf]

SNIP

Hopefully the lens on the camera doesn't have much effect.

The combined imaging chain MTF is approximately the multiplication of the
modulation of individual imaging components. It deteriorates rapidly, the
weakest link causing the worst chain result.

However, the main thing is that this test doesn't say much
about the differences between high-end scanners.

The thing that surprises me is the big difference in scans on
the same model.

It just requires more rigid parameters (e.g. contrast and gamma), and more
samples per scanner to arrive at a more statistically significant average.

Bart

 

[Bart van der Wolf]

SNIP

I can do that and will send it in with my results.

Much appreciated, it will enhance understanding.

Just looked at the slide through my micrscope:

1/4 is still separated quite clearly, but much less so than 1/3. 1/5 is
very murky, extremely low contrast. 1/6 is still softly seperated in the
vertical orientation and there is an slight indication of an idea of
seperation in the horizontals, but I'ld rather not be put under oath for
this one. 2/1 is not seperated any more.

Do check. There seems to be a hint of horizontal (near vertical feature)
resolution in group 2, element 1 of
<http://www.jamesphotography.ca/bakeoff2004/minolta_DSE_5400_Bart_van_der_wo
lf_clip.jpg> so there ought to be something in the slide itself.

Looking forward to your contribution.

Bart

 

[Ralf C. Kohlrausch]

Do check. There seems to be a hint of horizontal (near vertical
feature) resolution in group 2, element 1 of
<http://www.jamesphotography.ca/bakeoff2004/minolta_DSE_5400_Bart_van_de
r_wo
lf_clip.jpg> so there ought to be something in the slide itself

Bad The slides seem to vary considerably in resulution. For some
reason Jim sent me two letters, so I have two slides. Now I checked the
second slide under my microscope and got resolution up to 2/2 in the
verticals and 2/1 in both orientations. That is exactly what I find on
your scan. Unsharp mask and enlargement brings it out mor clearly.

I'll ask my doctor to verify this, he allowed me to use his
laboratory-microscope for lens testing. It's an Olympus stereoscopic mic
that can go much higher an seemingly offers better contrast than mine.

I just tried a scan from the better slide and I can resolve one group
more than with the not so good slide.

If I use the 100%-viefinder of my Nikon F3 and frame a 20*30cm testchart
to the same enlargement as James I end up with a distance of about 75cm.
That is not the best working-distance for a 50mm standard lens.
Resolving 2/1 at this taking distance is a mere 60 lp/mm on the slide,
and extremely soft. A 2700-dpi-scanner could do around 55 lp/mm with a
high contrast testslide. That's what my Nikon LS-10 did, my Minolta
Quickscan 35 does about 49 lp/mm, but I have 118 lp/mm on the testslide.
I photographed a laserprinted (600dpi) USAF 1951 chart at 7.8 metres on
Sensia II 100. 1/6, 2/1 and 2/2 might also show limitations in the
printer-quality. At least it does with the 1200 dpi laserprints I have
here.

Greets
Ralf C.

 

[Jim Hutchison]

Bad The slides seem to vary considerably in resulution. For some
reason Jim sent me two letters, so I have two slides. Now I checked the
second slide under my microscope and got resolution up to 2/2 in the
verticals and 2/1 in both orientations. That is exactly what I find on
your scan. Unsharp mask and enlargement brings it out mor clearly.

I'll ask my doctor to verify this, he allowed me to use his
laboratory-microscope for lens testing. It's an Olympus stereoscopic mic
that can go much higher an seemingly offers better contrast than mine.

I just tried a scan from the better slide and I can resolve one group
more than with the not so good slide.

If I use the 100%-viefinder of my Nikon F3 and frame a 20*30cm testchart
to the same enlargement as James I end up with a distance of about 75cm.
That is not the best working-distance for a 50mm standard lens.
Resolving 2/1 at this taking distance is a mere 60 lp/mm on the slide,
and extremely soft. A 2700-dpi-scanner could do around 55 lp/mm with a
high contrast testslide. That's what my Nikon LS-10 did, my Minolta
Quickscan 35 does about 49 lp/mm, but I have 118 lp/mm on the testslide.
I photographed a laserprinted (600dpi) USAF 1951 chart at 7.8 metres on
Sensia II 100. 1/6, 2/1 and 2/2 might also show limitations in the
printer-quality. At least it does with the 1200 dpi laserprints I have
here.

Greets
Ralf C.

That *is* bad news. I shot 2 batches of slides, after running out of
the first, not realizing there would be this much interest. So which
slide is better - the first or the 2nd version? I'll send Marty
another one if it's the 2nd. If it's the first, then maybe you can
mail it to him.

I focused as best I could, but obviously they varied enough to make a
small - but perceptible - difference.

Realize that the difference is in fact only perceptible by high-end
scanners, so I'd like to address that by standardizing the results
with the same slide.

Additionally, is it possible the difference between these higher-end
scanners doesn't make as much a perceptible difference in the
end-product, as does comparing a Minolta 5400 with an Epson flatbed?
I'm wondering if telling the difference between an MTF50 rating of 29
and 31 is that valuable in the real world in terms of viewing images
printed from these scans?

Is it analogous to 2 competing students at adjacent positions at the
top of the bell curve? ... Just a thought.





jim h


~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

http://www.jamesphotography.ca

-free downloads
-scanning service

Even a bit of humour...

 

[Philip Homburg]

Additionally, is it possible the difference between these higher-end
scanners doesn't make as much a perceptible difference in the
end-product, as does comparing a Minolta 5400 with an Epson flatbed?
I'm wondering if telling the difference between an MTF50 rating of 29
and 31 is that valuable in the real world in terms of viewing images
printed from these scans?

Maybe it is an interesting test to shoot some Velvia 100F as good as possible
and send those slides around to people with high-end scanners.

Velvia is supposed to do 40 lp/mm at 1.6:1 compared to 30 lp/mm for Provia
100F. The interesting question this may answer is whether the film is the
limiting factor or not. If the LS-8000, LS-5000 and Minolta 5400 go up
by more or less an equal amount, then you may need sharp b/w film to really
get to the limit of those scanners. Furthermore, if the resolution goes
up enough, it 'proves' that your camera, lens and target, etc. are not
limiting factors in this test.

On the other hand, I am more worried about the big difference between scans
on same the model. Apparently, 'scanned on scanner brand/model' does tell the
whole story.

 

[Ralf C. Kohlrausch]

On the other hand, I am more worried about the big difference between
scans on same the model. Apparently, 'scanned on scanner brand/model'
does tell the whole story.

Hi,

please excuse me, English isn't my mothertongue. Should this be
_doesn't_ tell the whole story?

Greets
Ralf C.

 

[Ralf C. Kohlrausch]

The interesting question this may answer is whether the
film is the limiting factor or not. If the LS-8000, LS-5000 and
Minolta 5400 go up by more or less an equal amount, then you may need
sharp b/w film to really get to the limit of those scanners.
Furthermore, if the resolution goes up enough, it 'proves' that your
camera, lens and target, etc. are not limiting factors in this test.

Hi,

the film ist not the limiting factor, unless of course you skimp on
quality when shopping. Resolutiontests ar usually done at a contrast of
1:1000, which is virtually impossible to reach in a hombrew-test. If you
lighten a testchart at an angle of 45 degrees you my reach 1:8 or 1:16,
so resolution on the film should be visibly lower than in a professional
test. On the other hand, if you use the best f-stop (may vary 'tween
center and corner) at the best distance (normally infinity, or as close
to that as possible), focus manually on the screen with an
eyepiece-loupe, have your camera's focusing checked for alignment etc.
you can go well beyond 100 lp/mm even with fairly coarse grained Fuji
Sensia. But contrast gets low. The scanner then has a hard time getting
the fine but low contrast detail from the film. With better lenses and
films 200 lp/mm can reached - and surpassed. The German Fotomagazin
tests filmresolution by shooting testpatterns and then enlarging the
pics. They got a filmresolution of 200 lp/mm out again on the prints. So
there must have been even mor on the slide. In everyday-situations a
contrast of 1:1000 is easily possible. Imagine fine backlit branches of
a tree against the sky.

With lighting as it is at our tests will will learn more about the
scanners' ability of getting low contrast-detail than fine detail, if it
comes to the resolution wonders like Minolta 5400. Interesting enough, I
think.

Greets

Ralf C.

 

[Ralf C. Kohlrausch]

On Sun, 30 May 2004 00:30:58 +0200, "Ralf C. Kohlrausch" >

That *is* bad news. I shot 2 batches of slides, after running out of
the first, not realizing there would be this much interest. So which
slide is better - the first or the 2nd version? I'll send Marty
another one if it's the 2nd. If it's the first, then maybe you can
mail it to him.

Hi,

I received both letters the same day, both slides are marked 05/05/2004.
One is frame 19, one is frame 30. Did you shoot both batches the same
day, or do we have sample-variation within the batches? Of course I can
pass on the better slide. I'ld need the snailmail address. 30 is the
good one.

I focused as best I could, but obviously they varied enough to make a
small - but perceptible - difference.

Focusing can easily cut your resolution in half. I tried some lenstests
of my own and have ended up bracketing the focusing with a slider that
moves the camera by small increments. You should use 3mm-steps at a
distance of about 2,5 metres, bracketing at least two shots plus and two
shots minus. Of course that will be an awful lot of film :-(

Realize that the difference is in fact only perceptible by high-end
scanners,

No, it isn't. Even the 2700-dpi-class is affected, if at a lower level.
They will loose one or two grous, one gropu with my old Minolta
Quickscan 35.

so I'd like to address that by standardizing the results
with the same slide.

Additionally, is it possible the difference between these higher-end
scanners doesn't make as much a perceptible difference in the
end-product, as does comparing a Minolta 5400 with an Epson flatbed?
I'm wondering if telling the difference between an MTF50 rating of 29
and 31 is that valuable in the real world in terms of viewing images
printed from these scans?

The German computer magazine c't has tested scanners in their latest
edition. The Epson 4870 came out with 1650 dpi, far less than the
nominal 4800*9600.

Is it analogous to 2 competing students at adjacent positions at the
top of the bell curve? ... Just a thought.

I don't understand that one ;-)

Grrets
Ralf C.

 

[Philip Homburg]

Hi,

please excuse me, English isn't my mothertongue. Should this be
_doesn't_ tell the whole story?

Sorry. You are right.

 

[Philip Homburg]

On the other hand, if you use the best f-stop (may vary 'tween
center and corner) at the best distance (normally infinity, or as close
to that as possible), focus manually on the screen with an
eyepiece-loupe, have your camera's focusing checked for alignment etc.
you can go well beyond 100 lp/mm even with fairly coarse grained Fuji
Sensia. But contrast gets low.

With better lenses and
films 200 lp/mm can reached - and surpassed. The German Fotomagazin
tests filmresolution by shooting testpatterns and then enlarging the
pics. They got a filmresolution of 200 lp/mm out again on the prints. So
there must have been even mor on the slide. In everyday-situations a
contrast of 1:1000 is easily possible. Imagine fine backlit branches of
a tree against the sky.

That's why 'we' are looking at the MTF50 (and MTF20) values. Very
high contrast subjects are interesting, but the subjective sharpness of a
print is determined to a large extent by losing a little as possible
from low contrast subjects.

Hmm, Fuji specifies 70 lp/mm at 1000:1 for Provia 100F, and 67.5 lp/mm
for Sensia 100 (again at 1000:1).

Are you sure it is 200 lp/mm?

With lighting as it is at our tests will will learn more about the
scanners' ability of getting low contrast-detail than fine detail, if it
comes to the resolution wonders like Minolta 5400. Interesting enough, I
think.

My own (limited) experience with big prints suggests that low-contrast
details are more important then a few high-contrast details.

 

[Alan Browne]

...at http://www.jamesphotography.ca.


pls read my disclaimers before ripping my head off.....

I want to se the DSE 5400 with the Minolta s/w (that I did) v.
the DSE 5400 that Bart (et al) did with VueScan.

(See my second e-mail today (May 31) to you for link to the .zip).

Cheers,
Alan.

 

[Ralf C. Kohlrausch]

Hmm, Fuji specifies 70 lp/mm at 1000:1 for Provia 100F, and 67.5 lp/mm
for Sensia 100 (again at 1000:1).

Are you sure it is 200 lp/mm?

That's been a Kodak E 100 G.

Greets
Ralf C.

 

[Gary L Hunt]

I keep hearing about resolving grain, but I've never seen my LS5000 ED
resolve grain even at 4000 dpi. I'll not argue there's grain in the
images, but the only grain I find comes from the scanner. It's even
and smoothly distributed. ASA 400 film should show grain clumps,
strings, and all kinds of crappy looking stuff when blown way up. I
just don't see that.

Roger Halstead (K8RI & ARRL life member)
(N833R, S# CD-2 Worlds oldest Debonair)
www.rogerhalstead.com

I've read some of the "optical theory" debates on grain aliasing etc. on
this forum, and I don't claim to follow them. But, while I won't claim that
I'm seeing grain either, both my 4000ED and 500ED scanners show something
that looks like grain effects. As you say, it tends to be evenly-distributed
(and I don't use ISO 400 film, so I can't argue with you there.) But it does
vary a lot from one batch of film to the next. I've scanned 22,400 slides in
the past 6 weeks, and there is a significant difference in this "apparent grain"
depending on film type (and processing--it's much more prominent in my old
home-processed Ektachrome than it is in recent batches done by Fuji, and less
visible yet in the Kodachrome 64.) I don't see these differences being due
just to the scanner.

Gary Hunt

 

[Kennedy McEwen]

That's been a Kodak E 100 G.

In which case it certainly wasn't 200lp/mm, no matter what the contrast
was.
http://www.kodak.com/global/en/professional/support/techPubs/e4024/e4024.
pdf
shows the MTF of Kodak Ektachrome 100G (just the film, no camera lens
component included!) has fallen to an average of 18% (green) at only
80lp/mm - so you might get 200 *lines*/mm (ie. 100 line pairs/mm) if you
use a high contrast back illuminated test target (which saturates both
shadows and highlights) and the finest lens operating at its sweet spot.
200lp/mm is just kidding yourself - or a misprint!

The MTF50 figure used in this test is only 40lp/mm for E-100G, which is
comparable with a 2000ppi sampling density, assuming 100% fill factor.

In fact, if it was only film and scanner CCD that were in play then you
should distribute the MTF evenly between the two, resulting in 70% MTF
being the criteria for both components, resulting in the product being
50% MTF. In that case, you are looking at something closer to 25lp/mm,
which is comparable with only a 1400ppi ideal sampling sensor.

Of course, there are optics and residual scanner motion and a few other
things that come in to play as well, but film resolution is certainly
not as insignificant in these tests as you suggest - not by a long way.
Of course, MTF50 is a long way from the resolution limit of any system
as well.

 

[Ralf C. Kohlrausch]

Not exactly. Fuji specifies 135 and 140 lines/mm, which refers to the
black lines. I count the white lines as well and call an white and a
black line a linepair. You could use the term cycles as well. I got 118
linepairs/mm with a frontlit 1:8 to 1: 16 selfprinted testchart on
sensia II 100. I did that myself ;-)
Of course not, I didn't do it myself ;-)

In which case it certainly wasn't 200lp/mm, no matter what the
contrast was.
http://www.kodak.com/global/en/professional/support/techPubs/e4024/e4024
..
pdf
shows the MTF of Kodak Ektachrome 100G (just the film, no camera lens
component included!) has fallen to an average of 18% (green) at only
80lp/mm - so you might get 200 *lines*/mm (ie. 100 line pairs/mm) if
you use a high contrast back illuminated test target (which saturates
both shadows and highlights) and the finest lens operating at its
sweet spot. 200lp/mm is just kidding yourself - or a misprint!

well, I myself took the E 100 G to 94 linepairs/mm using an old MF
Nikkor 1,4/50 (the first edition of this lens) with a frontlit (1:8 to
1:16) selfprinted target. Fotomagazin uses a Leica 2,8/100 macro lens
and a backlit 1:1000 target

The MTF50 figure used in this test is only 40lp/mm for E-100G, which
is comparable with a 2000ppi sampling density, assuming 100% fill
factor.

In fact, if it was only film and scanner CCD that were in play then
you should distribute the MTF evenly between the two, resulting in
70% MTF being the criteria for both components, resulting in the
product being 50% MTF. In that case, you are looking at something
closer to 25lp/mm, which is comparable with only a 1400ppi ideal
sampling sensor.

Of course, there are optics and residual scanner motion and a few
other things that come in to play as well, but film resolution is
certainly not as insignificant in these tests as you suggest - not by
a long way. Of course, MTF50 is a long way from the resolution limit
of any system as well.

Well, my Nikon LS 10 (2700 dpi) my Minolta quickscan 35 (2820 dpi) and a
Fuji Frontier minilab (2700 dpi, if I am informed correctly) resolve 55
lp/mm, 49+ and 49+ of that abovementioned slide with 118 lp/mm. Your
theory may be good, but in real life I get about twice your numbers.

Greets
Ralf C.

 

[Philip Homburg]

Well, my Nikon LS 10 (2700 dpi) my Minolta quickscan 35 (2820 dpi) and a
Fuji Frontier minilab (2700 dpi, if I am informed correctly) resolve 55
lp/mm, 49+ and 49+ of that abovementioned slide with 118 lp/mm. Your
theory may be good, but in real life I get about twice your numbers.

Bart van der Wolf lists 45.4 lp/mm for an LS-2000. You are saying that an
LS-10 gets 55 lp/mm from a medium to low contrast slide?

 

[Fernando]

Well, my Nikon LS 10 (2700 dpi) my Minolta quickscan 35 (2820 dpi) and a

Fuji Frontier minilab (2700 dpi, if I am informed correctly) resolve 55
lp/mm, 49+ and 49+ of that abovementioned slide with 118 lp/mm.

This confirms my findings. On a lower res slide, my (now gone) Minolta
Scan Dual III resolved about 50 lp/mm as well:

http://gundam.srd.it/PhotoPages/epson_vs_minolta.html

My SS120 did about the same. :-(

Fer

 

[Ralf C. Kohlrausch]

Bart van der Wolf lists 45.4 lp/mm for an LS-2000. You are saying
that an LS-10 gets 55 lp/mm from a medium to low contrast slide?

I don't know Bart's slide, I know mine. As we have seen with the
bakeoff-slide even with careful focusing there may be tolerances that
reduce resolution. And yes, I am saying that a LS-10 gets 55 lp/mm from
a good quality slide.

Greets
Ralf C.