Targetscanning Vuescan and calibrating

  • Thread starter Thread starter I. Ordejon
  • Start date Start date
I

I. Ordejon

Hello everybody,

I know this has been discussed a couple of times but I am still not sure
about the proceedings.

I have a Wolf Faust target which I want to use with Vuescan and littlecms in
order to calibrate my scanner (Epson 3200). I have scanned the target at
3200dpi, 48bit inout and 48bit .tif output. No color settings ("None" in
vuescan) but the color space setting at "Adobe RGB (1998)". However I am not
at all happy with the results. Should I rather scan the target as "RAW" and
set the color space at "default device" (I suppose that the gamma is "1"
whenever the color setting is "none")?

I know that vuiescan has another method of calibrating but I just wanted to
compare with littlecms.

Thanks!

Ignacio
 
Won't anybody give it a try ;-) ? Ed? Bart? I have tried Vuescans own
calibration workflow and get better results than using the littlecms one
(hooray for VS!). Is VS calibration just "better" or am I missing something
(see my question above)?

Thanks,

Ignacio
 
I. Ordejon said:
Won't anybody give it a try ;-) ? Ed? Bart? I have tried Vuescans own
calibration workflow and get better results than using the littlecms one
(hooray for VS!). Is VS calibration just "better" or am I missing something
(see my question above)?

I read the original message where you said you wanted to try LCMS.
I have little to comment on LCMS, other than that it seems to do a
reasonable job for many who have tried it. If I remember correctly, it
produces theoretically more accurate profiles because they are based
on TRCs instead of only Primaries. That should be the main difference,
because AFAIK VueScan also uses the LCMS engine. VueScan uses
Primaries in a matrix, LCMS uses curves as LUT, for intermediate
colors.

VueScan is good enough to get you in better shape than without
profiling and it does it fast. Whether an additional effort is
beneficial, depends on your requirements.

Bart
 
Thanks, Bart, sorry for bothering :-) But, is any target profiling (apart
from VS own) finally based on a raw scan or on a tiff scan with no color
correction? How do I avoid any color correction in VS in order to provide
another profiling system with "untouched" data?

Regards,
Ignacio
 
Thanks, Bart, sorry for bothering :-) But, is any target profiling (apart
from VS own) finally based on a raw scan or on a tiff scan with no color
correction? How do I avoid any color correction in VS in order to provide
another profiling system with "untouched" data?

From what I've experienced, "scanning it right" is more difficult than
it seems.

1) Getting the right illumination.
This is not so trivial.
The target was meant to be scanned in "optimal light conditions"
(about 5500K, uniform, non-flaring). Some scanners have light
source-related problems, such as dark corners. If the scanning
software won't account for that (i.e., calibration), you could get
some patches darker than they should be, not because the scanner has
linearity issues but because of the dark corners...

2) Getting the right IT8 grid shape.
Often the target slide won't be perfectly "squared off". You'll have
to adjust the IT8 grid to have the measuring areas right on the
patches. Vuescan, for example, won't allow for any non-parallel
stretching (while LCMS profiler will allow any kind of stretching), so
you can't compensate for a slightly rotated slide, for example.
Yes, there's some tolerance 'cause only the centers of the areas are
measured, but still, if the film carrier is poorly designed, you can
have troubles having a good match.

3) Getting the right exposure
That's quite difficult.
Each scanner has its way to meter the exposure. The target has some
light (even nearly-white, but not *perfectly* transparent) patches and
some dark (even nearly-black, but not *fully* opaque) patches.
It's not trivial to have a good exposure, that is needed to have
"good" black patches and "good" white patches.
I found that often, I have troubles getting "good whites" and "good
blacks" at the same time. This affects profile building. I often scan
the target at various fixed exposures, try to have a good monitor
match (against the light table), select the "best" scans, and build
various profiles against those ones. Then, I check the profiles to
select the better one.
If you have an underexposed target scan, for example, the profile will
raise the luminosity of all your scans: not quite what you'd want.

4) Getting the right histogram.
The Color Balance setting in Vuescan is difficult to understand
completely, for me at least. What's the difference between "none" and
"neutral", for example? You could say it's the black and white points,
but it isn't: if you set both to 0%, still Neutral is different from
None.
"Neutral" seems to boost contrast a bit (even with both points at 0%),
and it seems to help having a better target scan (black and white
patches are "more black" and "more white", so nearer to the expected
reference IT8 values). But you see, it's not very clear how this
works...

5) Getting the right output color space, including gamma point and
white point (light source) setting.
Again, this is not trivial, for me at least.
We don't know exactly how the profiler that's embedded into Vuescan
works, but at least Vuescan knows it :) ; what about profiling with an
external tool like LCMS? Better to have the TIFF in AdobeRGB or in
sRGB? And does it matter at all? Better a 6500K white point as it's
the more widely used (and likely the one you'll process your images
with), or a 5000-5500K one (the one that the target vendor measured
with)?
Better a linear 1.0 gamma or a typical 2.2 one?

So you see, I'm quite confused too.
Let's hope we get enlighting answers by knowledgeable people (and
hopefully, from Ed Hamrick and Wolf Faust, too!). :)

All that said, I had good results with Vuescan profiles: better than
LCMS ones, probably 'cause with LCMS you have so many variables (from
scanning parameters to profile calculation parameters) that I mess it
up more easily.
Another big advantage of the matrix-based profiles that Vuescan
produces, is that they are tolerant to exposure variations.
This is agood thing, since I always adjust the CCD exposure to have
the optimal value when scanning slides (ie., if a slide is uniformly
dark, I boost the exposure in order to get the maximum detail from the
shadows, and then re-adjust the histograms within Photoshop).
Exposure varations are not well accounted for, with a LUT-based
profile ala LCMS.

Fernando
 
3) Getting the right exposure
That's quite difficult.
Each scanner has its way to meter the exposure. The target has some
light (even nearly-white, but not *perfectly* transparent) patches and
some dark (even nearly-black, but not *fully* opaque) patches.
It's not trivial to have a good exposure, that is needed to have
"good" black patches and "good" white patches.

This is someting I have not counted for so far. But - of course. Once you
think about it, even the exposure time has necessarily have to have some
(and not the least) influence on the profile.
4) Getting the right histogram.
The Color Balance setting in Vuescan is difficult to understand
completely, for me at least. What's the difference between "none" and
"neutral", for example?

Good question. Thats something I qould really like to know. I always
thought, that "none" was less than "neutral". But, who knows ...
5) Getting the right output color space, including gamma point and
white point (light source) setting.
Again, this is not trivial, for me at least.
We don't know exactly how the profiler that's embedded into Vuescan
works, but at least Vuescan knows it :) ; what about profiling with an
external tool like LCMS? Better to have the TIFF in AdobeRGB or in
sRGB? And does it matter at all? Better a 6500K white point as it's
the more widely used (and likely the one you'll process your images
with), or a 5000-5500K one (the one that the target vendor measured
with)?
Better a linear 1.0 gamma or a typical 2.2 one?

Exactly. There must be an ideal workflow for this - I hope :-)

All that said, I had good results with Vuescan profiles: better than
LCMS ones, probably 'cause with LCMS you have so many variables (from
scanning parameters to profile calculation parameters) that I mess it
up more easily.

Me too. And thats the strange thing. My Vuescan profiles simply work. My
lcms profiles seem to boost areas so unnaturally it hurts. I just would like
to know why :-)

Regards,
Ignacio
 
I. Ordejon said:
Thanks, Bart, sorry for bothering :-) But, is any target profiling (apart
from VS own) finally based on a raw scan or on a tiff scan with no color
correction?

VueScan bases its calibration on the Raw data. This means that neither
color balancing nor gamma (brightness) nor Black/White points settings
have any influence on the profile. I consider that a good approach.
All that needs to be done, is to get the exposure level close to the
level that will be used for other scans.

Many other profilers, behave better when they get to work with a gamma
adjusted image. I can imagine that especially 8-bit/channel
applications get better results that way, although there may be issues
if a profiling gamma adjustment is different from the final scan.
How do I avoid any color correction in VS in order to provide
another profiling system with "untouched" data?

VueScan does that (avoiding) automatically when set to profiling,
because it use Raw data. If you save a Raw file, you'll get the same,
and if you save as 24-bit Raw then a gamma adjustment will
automatically be applied. All very straight forward.

Bart
 
SNIP
4) Getting the right histogram.
The Color Balance setting in Vuescan is difficult to understand
completely, for me at least. What's the difference between "none" and
"neutral", for example? You could say it's the black and white points,
but it isn't: if you set both to 0%, still Neutral is different from
None.

'None' just gives what is scanned, no Black/White points are applied,
with the chosen gamma/brightness. That means that only a part of the
full 16-bit/channel range will be occupied with data.

'Neutral' will stretch/normalize the histogram data to fill the 16-bit
range from low to high (Black/White points set to 0 % clipping),
without changing the relative positions between the channels. That
means that one channel may be responsible for the lowest histogram bin
data, while another channel may be responsible for the highest bin
value and other channels may not reach to the extremes, just like with
'None'. That may (depending on the profiling software) lead to
Whitepoint clipping, because Whites are usually expected to be around
235 (in an 8-bit range).

So, if you don't want to reduce the Raw data to 8 b/ch by choosing a
24-bit Raw output file, you are probably better of by starting with
'None' for Color balancing, and by adjusting the exposure level so the
brightest whites are around 90-95% of maximum for the brightest
channel. The exposure level will ultimately be different for the
specular highlights of an image, but that depends on the workflow used
when scanning.

Bart
 
I. Ordejon said:
Won't anybody give it a try ;-) ? Ed? Bart? I have tried Vuescans own
calibration workflow and get better results than using the littlecms one
(hooray for VS!). Is VS calibration just "better" or am I missing
something
(see my question above)?

The reason VueScan works better is that the VueScan profiles work
independent of CCD exposure time, while TRC-based profiles (from LCMS)
only work with a fixed CCD exposure time.

That's why VueScan uses matrix profiles - so they'll work properly
when you vary CCD exposure time.

Regards,
Ed Hamrick
 
Thanks, Ed. This is good news. Taking into consideration, that exposure time
may vary, I suppose I would otherwise have to use different profiles for
different exposure times.

Thanks for all answers.

Ignacio
 
Back
Top