MTF test: Epson 2450 vs. 4180

I was not seeing any practical difference between my old 2450 and the
new 4180; so I tried an Imatest comparison between the two.

I published 4180 results here, some time ago; I'll repeat them side by
side with the 2450 figures I just got.

Same target (a SET target: framed razor blade at 5°). Same software
(Vuescan professional), same settings (max hardware resolution, 48bpp,
no sharpening, no filters, no white balance nor any other histogram
adjustments), same crop location (around the center of the blade).
I instructed Imatest to take care of the different scanning resolution
(2400 dpi vs. 4800) when computing the cy/mm figures; so those figures
are rightly comparable.

2450@2400dpi (4180@4800dpi) MTF test
==================

-Resolution:
10-90% rise = 4.5 pixels (7.84)
MTF@50 = 12.8 cy/mm (13.3)
MTF@30 = 18.2 cy/mm (20.1)
MTF@Nyquist = 0.0272 (0.0079)

-CA:
CA Area = 0.285 pixels (0.234)
CA Crossing = 0.237 pixels (0.284)


Pretty small difference indeed; in my opinion, at least.

Fernando

"Fernando" <fcarello@tiscalinet.it> wrote in message
news:fm6lu0h6pj3acf9o0d11bid05ki39sh9ap@4ax.com...
SNIP
Just to make sure I understand what you did;

> I instructed Imatest to take care of the different scanning
> resolution (2400 dpi vs. 4800) when computing the cy/mm
> figures; so those figures are rightly comparable.

Did you instruct Imatest to interpret data as 2400 ppi, respectively
4800 ppi, or did you use the same value for both sampling densities?

> 2450@2400dpi (4180@4800dpi) MTF test
> ==================
>
> -Resolution:
> 10-90% rise = 4.5 pixels (7.84)
> MTF@50 = 12.8 cy/mm (13.3)
> MTF@30 = 18.2 cy/mm (20.1)
> MTF@Nyquist = 0.0272 (0.0079)
>
> -CA:
> CA Area = 0.285 pixels (0.234)
> CA Crossing = 0.237 pixels (0.284)

Bart

>> I instructed Imatest to take care of the different scanning
>> resolution (2400 dpi vs. 4800) when computing the cy/mm
>> figures; so those figures are rightly comparable.
>
>Did you instruct Imatest to interpret data as 2400 ppi, respectively
>4800 ppi, or did you use the same value for both sampling densities?

The first interpretation is the correct one.

So, (at 4800 dpi) the 4180 only gains about 10% more linear
resolution.
At 2400dpi, figures are virtually identical between the 4180 and the
2450, that goes to show that no improvements were made to the optics,
the glass, the microlenses, whatever.
BTW, I did not post the LPH/LPW figures 'cause I did not compute them
for the 4180; and I even accidentally erase the original scan.
I should re-run the test from scratch to get them, but anyway, I think
the story is pretty evident.

Now there's the 4990... once again, I don't think we'll see resolution
improvements.
This means I'll have to shell out for that LS9000 in the end!! :'(

Fernando

"Fernando" <fcarello@tiscalinet.it> wrote in message
news:4prlu0doo3frbabqn7brskhua58utol7cg@4ax.com...
>>> I instructed Imatest to take care of the different scanning
>>> resolution (2400 dpi vs. 4800) when computing the cy/mm
>>> figures; so those figures are rightly comparable.
>>
>>Did you instruct Imatest to interpret data as 2400 ppi, respectively
>>4800 ppi, or did you use the same value for both sampling densities?
>
> The first interpretation is the correct one.

That's what I thought (and thus you did it the correct way), but what
I wanted to be sure of.

> So, (at 4800 dpi) the 4180 only gains about 10% more linear
> resolution.

In absolute numbers, yes.

> At 2400dpi, figures are virtually identical between the 4180 and
> the 2450, that goes to show that no improvements were made
> to the optics, the glass, the microlenses, whatever.

Yes, despite the 2x higher sampling density of a near perfect
testobject, the resolution is apparently "lens limited", although the
4180 does extract a bit more detail form the same input compared to
the 2450. Part of the limitation is caused by a fixed focus lens
design an additional glass platen in the optical path (=two additional
uncoated air/glass surfaces).

> BTW, I did not post the LPH/LPW figures 'cause I did not
> compute them for the 4180; and I even accidentally erase the
> original scan. I should re-run the test from scratch to get
> them, but anyway, I think the story is pretty evident.

You don't really have to re-run them, just (correctly) assume twice as
high a number of output pixels for the 4180, which requires half the
equal output size magnification.
Because you can produce the same size output with half the
magnification for the 4180, you have effectively an almost twice
higher (but still limited) output(!) resolution.

So if the 2450 output is interpolated up to the same output pixels
size as the 4180 output, the output resolution would become:
10-90% rise = 9.0 pixels vs 7.84pixels
MTF@50 = 6.4 cy/mm vs 13.3 cy/mm
MTF@30 = 9.1 cy/mm 20.1 cy/mm
Which is made clear with the LPPH metric, or if you plot the ".csv"
output. It also makes clear that output, especially from the 4180, can
be down-sampled (if done properly) with relatively little loss, but
with the benefit of reduced (size/visibility) noise for archiving.

> Now there's the 4990... once again, I don't think we'll see
> resolution improvements.

QED !;-)

> This means I'll have to shell out for that LS9000 in the end!!
> :'(

It's most likely the better choice for existing film larger than 35mm
and up to MF, but the current scanner's price will also allow to
consider a digital camera (DSLR) for future shots. My goal has been to
postpone the transition as long as my budget allows, but it is
inevitable in the end (the timing being depending on one's current
hardware in use).

Bart

On Mon, 17 Jan 2005 02:39:34 +0100, "Bart van der Wolf"
<bvdwolf@no.spam> wrote:

>Yes, despite the 2x higher sampling density of a near perfect
>testobject, the resolution is apparently "lens limited", although the
>4180 does extract a bit more detail form the same input compared to
>the 2450. Part of the limitation is caused by a fixed focus lens
>design an additional glass platen in the optical path (=two additional
>uncoated air/glass surfaces).

Yes; and probably to get some DOF (needed to account for lack of AF)
the lens uses a small aperture, leading to resolution loss due to
diffraction.
I'd really love to test some good flatbed.
There are some models that actually move the lens/lensor across both
axis, so taking some sort of "grid scan". On some models, the optics
actually auto-focus. They should be awesome for large panoramic
frames, where dedicated filmscanners cannot help.

>> This means I'll have to shell out for that LS9000 in the end!!
>> :'(
>
>It's most likely the better choice for existing film larger than 35mm
>and up to MF, but the current scanner's price will also allow to
>consider a digital camera (DSLR) for future shots. My goal has been to
>postpone the transition as long as my budget allows, but it is
>inevitable in the end (the timing being depending on one's current
>hardware in use).

I already have a 10D, that I use for some low-light work. At 1600 iso
you still get useable colour images, and that's really unmatched by
film (I tried very hard).
But when it comes to sheer resolving power under good conditions,
nothing short of a 12mp camera is a match for good scanned 35mm film,
let alone 120. I did both test chart comparison and real-world scenes
comparison, and Astia/Velvia/Provia scans simply blow my 10D out of
the water.
I think I'll wait until 12-15mp cameras will get under $2000 and will
severely shrink in size and weight (1D/1Ds are benemoths! I'd rather
carry my MF gear) to get rid of my 35mm gear; but MF will stay longer
in my arsenal. :-)

Fernando

PS: still streaks and low DMax from 8.1.24? A bit better here, but
still not on par with M.S.U.