Epson 4180: Imatest run

[Fernando]

Hi all,
here at work we got a new Epson 4180, and I'm doing some Imatest run on
it (Imatest v.1.2.5) using Vuescan 8.1.19.

I hope someone may found those figures interesting.

Slanted Edge Test (framed razor blade at 6 degrees) at 4800dpi, no
histogram clipping (RGB exposure = 1.0), color balance to "None".
No sharpening, and I will address this point later on.

Cy/mm:
10-90% rise = 7.84 pixels
MTF@50 = 13.3 cy/mm
MTF@30 = 20.1 cy/mm
MTF@Nyquist = 0.0079

CA:
CA Area = 0.234 pixels
CA Crossing = 0.284 pixels

Shannon capacity:
@100% contrast = 1.36 bits/pixel = 0.0792 MB


My *personal* considerations: this scanner has very similar (and quite
low overall) resolving power of my older 2450. It's severely
lens-limited. No matter how high the resolution of the CCD sensor the
guy at Epson put in their scanners, they should really try to improve
the optics...

That said, an important point is sharpening. Due to the peculiar
geometry of Epson's Hyper-CCD sensor array, the scans from their
scanners need a larger amount of sharpening than other models.
Imatest usually shows both unsharpened and "standard-sharpened"
figures, I wonder which sharpening radius could do justice to the
Hyper-CCD (I'll post the new figures after having clarified this
point)?

Bart, Kennedy, any suggestion?

I'll test the scanner at 2400 dpi soon.
Please don't shoot the messenger...

Fernando

 

[David J. Littleboy]

Cy/mm:
10-90% rise = 7.84 pixels
MTF@50 = 13.3 cy/mm
MTF@30 = 20.1 cy/mm
MTF@Nyquist = 0.0079

Is that as bad as I thing it is? Well under 1/2 of what the 4000 dpi real
film scanners cough up. No? Yes????

Please don't shoot the messenger...

BANG!!!!!!!!!

FWIW, this guy's 4870 scans don't look half as bad as what you and Roland
(was it?) are/were getting. Maybe there are gross sample-to-sample
variations. (Yes, I realize that you're looking at the 4180, but I'm
assuming that uses the same imaging system as the 4870.)

http://www.gnyman.com/Personal/Epson3200vsEpson4870vsCanon9900F.htm

Also, there's the question of using Kami fluid with the 4870 that remains
untried.

David J. Littleboy
Tokyo, Japan

 

[Fernando]

I don't know about Gnyman... you see, same size/magnification crops
that is strange enough (being from 3200, 4800 and 4000 dpi scanners),
11MB image used as the full-size source and it's a 6x7 original (I get
340 MB scanning 6x4.5 on my Polaroid), 160ISO neg film which has
nowhere the ability to resolve of a modern 100iso slide, inconsistent
results (just look at the second series of crops -red cliff rocks-,
where the 8000 puts the 4870 to shame, compared to the other series of
crops), and so on.
This is the purpose of tests conducted in a more technical way (with a
SET target ad Imatest): to avoid those obscure variables as much as
possibile.

Yes, I think that, just as my 2450, this new 4180 is barely worth for
4x5" scans, and even then, it fails to do justice to LF lenses that, as
lower-resolution as you want, are nonetheless capable of well over 50
lp/mm on film under good conditions. A spatial frequency that this
scanner simply can't read in a useful way.

By the way, 2400dpi test results (same conditions as above; I still
have to find a proper sharpening setting, so there are unsharpened
runs):

4180 at 2400 DPI
======
Cy/mm:
10-90% rise = 4.47 pixels
M.T.F.@50 = 12.2 cy/mm
M.T.F.@30 = 18.4 cy/mm
M.T.F.@Nyquist = 0.0092
CA:
CA Area = 0.209 pixels
CA Crossing = 0.198 pixels
Shannon capacity:
At 100% contrast = 2.85 bits/pixel = 0.0528 MB

 

[Bart van der Wolf]

Hi all,
here at work we got a new Epson 4180, and I'm doing some Imatest
run on it (Imatest v.1.2.5) using Vuescan 8.1.19.

I hope someone may found those figures interesting.

Slanted Edge Test (framed razor blade at 6 degrees) at 4800dpi, no
histogram clipping (RGB exposure = 1.0), color balance to "None".

Just as a suggestion to improve consistency, you could consider saving
as a Raw file and evaluate in Imatest at gamma=1.0. This is just to
eliminate any influence of actual gamma adjustment being different
than instructed or color balance being applied. Just try and avoid the
clipping, as you do.

No sharpening, and I will address this point later on.

Cy/mm:
10-90% rise = 7.84 pixels
MTF@50 = 13.3 cy/mm
MTF@30 = 20.1 cy/mm
MTF@Nyquist = 0.0079

Okay, looks familiar (based on my 2450 experience). However, do
realise that a LPPH figure is more meaningful if you want to compare
final output quality between models (see below).

CA:
CA Area = 0.234 pixels
CA Crossing = 0.284 pixels

If you evaluate Raws, then the CA may be less significant because the
individual channels do not necessarily produce neutral color balance
(because it hasn't yet been applied). However, the CA is only a
fraction of a pixel, so it looks decent enough.

Shannon capacity:
@100% contrast = 1.36 bits/pixel = 0.0792 MB


My *personal* considerations: this scanner has very similar
(and quite low overall) resolving power of my older 2450. It's
severely lens-limited. No matter how high the resolution of the
CCD sensor the guy at Epson put in their scanners, they
should really try to improve the optics...

True, with an MTF@Nyquist of 0.0079 it is clearly not CCD or sampling
density limited, but don't forget the differences in output size. The
2450 needs 2x more enlargement for the same size output compared to
the 4800 model. That means you must either divide the 4800's cy/mm
resolution by 2, or multiply the 2450's resolution in cy/mm by 2 if
you want to compare cy/mm between them.

The LPPH metric takes care of that automatically if you instruct
Imatest to use the theoretical maximum image size in pixels (to
compare at equal output magnification).
So, assuming a scan from 35mm film, e.g. 4800ppi x 24mm / 25.4mm =
4535px height, and 4800ppi x 36mm / 25.4mm = 6803px width for the 4800
model, and similarly 2268 x 3402px for the 2450 model. For MF scans
these numbers obviously go up proportionally.

Anyway, it also means you can down-sample the resulting files to, say,
1/4th of their linear size, without losing much resolution, while
maintaining great sharpening potential with little risk for aliasing.
Of course a 10x15cm maximum output size (good quality though) from a
35mm film or an 8x10in from a MF film is a bit underwhelming...

That said, an important point is sharpening. Due to the peculiar
geometry of Epson's Hyper-CCD sensor array, the scans from
their scanners need a larger amount of sharpening than other
models.

Correct, a larger radius must be used for a comparable effect. However
part of the abundence of pixels (almost empty resolution) can be used
to print at higher ppi settings with less need for interpolation (=
empty resolution). So one could down-sample the 4800ppi scans to
2400ppi, sharpen, and print at 600 or 720ppi. Sharpening should of
course be matched to final output size.

Imatest usually shows both unsharpened and "standard-
sharpened" figures, I wonder which sharpening radius could
do justice to the Hyper-CCD (I'll post the new figures after
having clarified this point)?

Bart, Kennedy, any suggestion?

Don't mistake the Imatest sharpening radius for USM sharpening radius.
The Imatest radius is IMHO only useful for leveling the playing field
between digicams that apply unknown, but potentially different,
in-camera sharpening which influences the apparent resolution.
Post-processing sharpening produces different effects than the Imatest
radius may suggest. It is easy to verify if you apply post-processing
sharpening to a slanted edge crop, and compare the 10-90% edge
transition in Imatest from different radii and amount settings.

For all the above, keep in mind that different films will produce
different combined MFT results than the scanner alone, although the
scanner is clearly the weakest link in the chain. There may also be
differences between scanners.

One of the fun things you can derive from using Imatest is, how does a
particular material (film or reflection copy) perform on a given
scanner, and how much sharpening can be applied without causing
visible halo, and how large an output can one make without visible
deterioration (assuming a perfected imaging chain). It can also
provide guidance to produce tolerable file sizes if storage space is a
concern. It's a very useful tool for perfecting the workflow by
pinpointing the weakest link (besides the limitation from financial
resources).

Bart

 

[Bart van der Wolf]

SNIP

FWIW, this guy's 4870 scans don't look half as bad as what
you and Roland (was it?) are/were getting. Maybe there are
gross sample-to-sample variations.

I think the major difference may be that the comparisons based on
cy/mm must be interpreted towards output size magnification, where
equal output size comparisons benefit from sheer numbers of pixels and
less magnification.

SNIP of interesting user review

Also, there's the question of using Kami fluid with the 4870
that remains untried.

It will definitively improve contrast and reduce visibility of
scratches. However, it may position the film in a slightly less then
ideal focus plane for fixed focus scanners, so it'll be a trade-off
(unless the fixed focus plane happens to coincide with the film
position on the platen surface).

Bart

 

[Fernando]

Just as a suggestion to improve consistency, you could consider saving
as a Raw file and evaluate in Imatest at gamma=1.0. This is just to
eliminate any influence of actual gamma adjustment being different
than instructed or color balance being applied. Just try and avoid the
clipping, as you do.

It is a good general advice; anyway, I run those tests on the same
machine, with the same software and the same settings. So when I see 60
cy/mm from the SE5400, 54 cy/mm from the SS120 and 13 cy/mm from the
4180... I start to wonder why people keep saying that flatbed scanners
are approaching dedicated film scanners (just look at photo-i.co.uk!).

True, with an MTF@Nyquist of 0.0079 it is clearly not CCD or sampling

density limited, but don't forget the differences in output size. The

2450 needs 2x more enlargement for the same size output compared to
the 4800 model. That means you must either divide the 4800's cy/mm
resolution by 2, or multiply the 2450's resolution in cy/mm by 2 if
you want to compare cy/mm between them.

I did another run with the 4180 set at 2400 (it's posted below), and
obtained more or less the same results (maybe 12.5 cy/mm instead of
13.3, or something like that). Practically, the 4180 gains almost
nothing from 2400 to 4800 dpi (and this concurs with our findings about
being lens-limited and so).

The LPPH metric takes care of that automatically if you instruct
Imatest to use the theoretical maximum image size in pixels (to
compare at equal output magnification).

Yes, but to a photographer that wants to scan film, a cy/mm figure
(quite similar to the well known lp/mm figure) tells more; or at least,
it tells more to me.
Because one can relate it to the figures from his lenses (I know that
my best MF lenses still have some contrast above 70 lp/mm, on optimal
conditions).
But you're right, my future tests will also provide LPPH figures.

Don't mistake the Imatest sharpening radius for USM sharpening radius.
The Imatest radius is IMHO only useful for leveling the playing field

between digicams that apply unknown, but potentially different,
in-camera sharpening which influences the apparent resolution.
Post-processing sharpening produces different effects than the Imatest
radius may suggest. It is easy to verify if you apply post-processing

sharpening to a slanted edge crop, and compare the 10-90% edge
transition in Imatest from different radii and amount settings.

Interesting.
So, to do justice to the HyperCCD, I should experiment with different
USM, applied before the Imatest runs? Choosing, in the end, the best
sharpening that does not show a visible "crest" in the MTF plot?
Thanks!

Fernando

 

[Fernando]

BANG!!!!!!!!!

I'm dead!! :'(

Also, there's the question of using Kami fluid with the 4870 that remains
untried.

Yes, this could be interesting.
Some time ago, a guy did a test on a 2450 with Kami fluid, and the
results were quite better; but he did not use the Kami fluid in a
closed film carrier as suggested; instead, he just poured the fluid
onto the scanner glass surface (!) and flattened the film on the wet
bed... this way he had an hard time cleaning the scanner, but he also
avoided the focus shift that a closed carrier would have brought.

One of the reasons I'd like to upgrade my scanners to a LS-9000 is
precisely to try Aztek's special film carrier that allows Kami fluid to
be used on the LS-8000/9000 units.

Fernando

 

[Bart van der Wolf]

SNIP

I start to wonder why people keep saying that flatbed
scanners are approaching dedicated film scanners (just
look at photo-i.co.uk!).

You won't hear me saying that
Maybe they say it because they have never seen a good alterative.

SNIP

Practically, the 4180 gains almost nothing from 2400 to
4800 dpi (and this concurs with our findings about being
lens-limited and so).

Although it could make a difference in noise performance. It is better
to over-sample and down-sample, than to exceed the Nyquist frequency
with significant modulation. But in this case the Nyquist frequency is
too far removed from the limits imposed by the lens to benefit much.
It is only a better alternative to upsizing, because that has even
less resolution.

SNIP

So, to do justice to the HyperCCD, I should experiment
with different USM, applied before the Imatest runs?
Choosing, in the end, the best sharpening that does not
show a visible "crest" in the MTF plot?

Yes, although it is easier to visually pinpoint the sweetspot on the
10-90% edge transition graph when you prepare a few cropped samples
from the original (using exactly the same spot from the original helps
accurate judgment). Just make sure you overwrite the default size
settings for an LPPH comparison. The SFR/MTF graph will then tell you
how the different frequencies have benefited. The lower frequencies
will change more than the higher frequencies with a regular USM, while
High Pass sharpening produces much more of an overall improvement.

A theoretical sharpening example, based on a "perfect" artificial
slanted edge with a Gaussian blur of radius 1.0 applied (produces a
10-90% rise of 2.91 pixels), is shown here:
- with USM: http://www.xs4all.nl/~bvdwolf/temp/USM.png
- with HPS: http://www.xs4all.nl/~bvdwolf/temp/HPS.png
This indicates that High Pass Sharpening (*HPS) improves the MTF but
also increases noise (if applied before Neat Image) and thus restores
capture losses better, while USM improves visual (output) contrast
stronger but produces halo when applied excessively.

* High Pass Sharpening can be applied in Photoshop by duplicating the
actual image layer, setting Blending mode to Overlay, and applying the
Filter|Other|High pass filter to it. Adding an Edge mask to it will
suppress noise amplification in featureless areas such as skies.
Multiple HPS layers can be applied.

Bart

 

[Fernando]

The lower frequencies
will change more than the higher frequencies with a regular USM, while
High Pass sharpening produces much more of an overall improvement.

Interesting. I actually use FocalBlade for my sharpening needs, but...
are you aware of any action/plugin that exploits HPS in a useful way?
Fernando

 

[Bart van der Wolf]

Interesting. I actually use FocalBlade for my sharpening needs,
but...
are you aware of any action/plugin that exploits HPS in a useful
way?

I made an action myself ;-) but you can try FocalBlade on a 16/15-bit
slanted edge made in Photoshop I used a 200x599 px target (to avoid
clipping I used 55 and 215 for the 8-bit gray values). If you Gaussian
Blur it with the same (Radius 1) amount as I did, then you can try
different FB settings to see if they are more effective.

The biggest issue for filmscans is to avoid sharpening the graininess,
which essentially means applying a noise reduction early in the
postprocessing stages.

There are probably commercial sharpening routines with fancy names
that are based on HPS and or USM techniques
(http://www.pixelgenius.com/sharpener/index.html), but with the
combination of Neatimage (which also has a frequency based sharpening
option) and one or two Photoshop actions (sharpening and edge mask) I
have lots of control.
<http://www.thelightsrightstudio.com/DigitalDarkroom/PhotoshopTools/TLRSharpeningToolkit.htm>
seems to address the most important issues too, but is free, although
I haven't studied it in depth yet.
<http://www.reindeergraphics.com/free.shtml#selectedge> may be useful
to incorporate in your own actions.

 

[R. Bailey]

Fernando penned:

I did another run with the 4180 set at 2400 (it's posted below), and
obtained more or less the same results (maybe 12.5 cy/mm instead of 13.3,
or something like that). Practically, the 4180 gains almost nothing from
2400 to 4800 dpi (and this concurs with our findings about being
lens-limited and so).

In light of this fact, would it be reasonable to to conjecture that the
Epson 2480, rated at 2400 dpi and a fraction of the cost of the 4180,
would be a better buy dollar-wise?

Thanks for providing your analysis, it is always enlightening to have
facts at hand.

 

[Fernando]

In light of this fact, would it be reasonable to to conjecture that the
Epson 2480, rated at 2400 dpi and a fraction of the cost of the 4180,
would be a better buy dollar-wise?

Hello!

Well, the 2480 is a more limited machine: for example, it only supports
35mm film (and only 2 slides at a time) while the 4180 also supports 120
film (and even panoramic 6x17 frames).
And maybe (I don't have the information right here), the film holder is
built into the cover for the 2480, while the 4180 uses a traditional
film holder to be laid on the glass bed. I prefer this solution, because
1) it's easier to clean the glass than the built-in film loader and 2)
you can always built your own film holder (for example from black thick
cardboard) and enjoy the whole 4180 transparency lid (76x270mm) for
custom purposes ("strange" film formats and so).

That said, if you only plan to scan 35mm, the 2480 has the best
price/performance ratio in my opinion.

Fernando

 

[R. Bailey]

Fernando penned:

Well, the 2480 is a more limited machine: for example, it only supports
35mm film (and only 2 slides at a time) while the 4180 also supports 120
film (and even panoramic 6x17 frames).

Hmm, I thought the 2480 had 6x6 holders, but maybe they were only 35mm
slide holders. The lit part of the lid looked big enough for 120, but
maybe not.

And maybe (I don't have the information right here), the film holder is
built into the cover for the 2480, while the 4180 uses a traditional
film holder to be laid on the glass bed. I prefer this solution, because

This is similar to the 2480 that I saw. Perhaps Epson has a variations in
the same model line?

That said, if you only plan to scan 35mm, the 2480 has the best
price/performance ratio in my opinion.

For 35mm, the good dedicated scanners are not so expensive as the 120
scanners, so it is easier to get a much better scanner for a bit more.

The 4180 seems to perform so poorly judging from your numbers. About what
effective print size could one make from a 6x6 frame scanned on a 4180?
Sharp 11x14s or 16x20s? Yes, no, maybe?

Thanks for your input.

 

[Fernando]

Hmm, I thought the 2480 had 6x6 holders, but maybe they were only 35mm
slide holders. The lit part of the lid looked big enough for 120, but
maybe not.

The 2480 has a transparency unit of only 49x125mm, not enough for 120
unfortunately (moreover, the TPUs tend to vignette at borders)!

The 4180 seems to perform so poorly judging from your numbers. About what
effective print size could one make from a 6x6 frame scanned on a 4180?
Sharp 11x14s or 16x20s? Yes, no, maybe?

I think you can obtain a sharp 11x11", and a fair 14x14", from a good
6x6 original, with careful unsharp masking.
Larger prints may be fair if viewed from some distance (I tend to put my
nose in front of prints, but that's me!).

Fernando

 

[Al]

Hi all,
here at work we got a new Epson 4180, and I'm doing some Imatest run on
it (Imatest v.1.2.5) using Vuescan 8.1.19.

I hope someone may found those figures interesting.

Slanted Edge Test (framed razor blade at 6 degrees) at 4800dpi, no
histogram clipping (RGB exposure = 1.0), color balance to "None".
No sharpening, and I will address this point later on.

Cy/mm:
10-90% rise = 7.84 pixels
MTF@50 = 13.3 cy/mm
MTF@30 = 20.1 cy/mm
MTF@Nyquist = 0.0079

CA:
CA Area = 0.234 pixels
CA Crossing = 0.284 pixels

Shannon capacity:
@100% contrast = 1.36 bits/pixel = 0.0792 MB


My *personal* considerations: this scanner has very similar (and quite
low overall) resolving power of my older 2450. It's severely
lens-limited. No matter how high the resolution of the CCD sensor the
guy at Epson put in their scanners, they should really try to improve
the optics...

I was trying to decide between the 4180 and the Microtek s400. This
makes the decision a little more difficult because sometimes the low
priced unknown seems like less of a risk than the known.

Maybe the Epson 3170 might be a good alternative for a few dollars
less with medium format capabilities.

My experience with Microtek is that they have trouble with color
registration - RGB are not aligned. It would be interesting to see a
test of one of their i or s series 4800 dpi scanners though.

 

[Fernando]

I was trying to decide between the 4180 and the Microtek s400. This
makes the decision a little more difficult because sometimes the low
priced unknown seems like less of a risk than the known.

Maybe the Epson 3170 might be a good alternative for a few dollars
less with medium format capabilities.

This could well be the case.
I would not have upgraded from the 2450 to the 4180, but I needed a
2nd scanner at work...
And from what I'm gathering, the 4870 and then the 4990 are no better
for what concerns the resolving power. Too bad.

My experience with Microtek is that they have trouble with color
registration - RGB are not aligned. It would be interesting to see a
test of one of their i or s series 4800 dpi scanners though.

My Polaroid SS120 (aka Microtek 120TF) has CCD registration problems,
too.
I keep sending it to the repair facility, and after a few months (just
after the repair warranty ends), it starts showing misregistration
again.
I'll buy an LS-9000 or a new Minolta Multi Pro if they upgrade it.

My advice is to beware of Microtek products.

Fernando