Creating an accurate scanner profile?

  • Thread starter Thread starter Patrick P.
  • Start date Start date
P

Patrick P.

Hello all,

I've purchased some IT8 targets for Provia and Kodak from Wolf Faust.
I've made a few profiles but am not completely happy with my results.
I've made profiles before with a Kodak Q60 target and got decent
results so I'm not sure if it is something I'm doing or maybe the
target/reference.

Here's what I've tried. I first scan the target on my Minolta 5400
with auto exp off, no color management, and no adjustment. I tried
making profiles with Little CMS, Diblook and Monaco EZ color. What I'm
finding is that the colors on the target from B13-B19 are almost
white. If you don't know what I'm talking about have a look at
http://www.targets.coloraid.de/ (B13-B19). I don't know why this range
isn't getting picked up (or getting whited out). I tried with scans
that were multisampled at x1, x4 and x16 but didn't find much
difference in the profiles.

Has anyone encountered this type of problem or have any idea how to
solve it?

On a side note: With little CMS should I be using the "patched
qtscanner profiler for film" or the original one that came in the zip
file (downloaded from http://www.littlecms.com/profilers.htm) to
produce the most accurate results? They seem to be very different in
file size and results.

Also with Little CMS is there a way to load a 16 bit Tiff file? If not
am I sacrificing accuracy by converting to another format like PNG and
8bit? What is suggested?

And lastly is there any trick to using Monaco EZ color? I can't seem
to find any way to get to a options screen when creating a profile.

I guess the question is what do I need to do to get the most accurate
profile?

Thanks,
Patrick
 
Patrick P. schrieb:

Patrick,
On a side note: With little CMS should I be using the "patched
qtscanner profiler for film" or the original one that came in the zip
file (downloaded from http://www.littlecms.com/profilers.htm) to
produce the most accurate results? They seem to be very different in
file size and results.
Click to expand...
probably rather the new Beta profiler at
http://www.littlecms.com/lcms_profiler_beta_3.exe

Btw, is your scanner color space an approx. linear
(gamma=1) one? At least the old, open-source LCMS
profiler had accuracy problems with linear scans.
Also with Little CMS is there a way to load a 16 bit Tiff file?
Click to expand...
Yes, with tifficc you can also apply profiles to 16bpp tiff
files (use the -w option, if you also want 16bpp output).

Regards,
Gerhard
 
Gerhard,
probably rather the new Beta profiler at
http://www.littlecms.com/lcms_profiler_beta_3.exe
Click to expand...

Thanks for the reply. I'll try the new beta version when I get home
from work.
Btw, is your scanner color space an approx. linear
(gamma=1) one? At least the old, open-source LCMS
profiler had accuracy problems with linear scans.
Click to expand...

Not sure how to tell. My scanner has 3 options, '8 bit', '16 bit' and
'16 bit linear'. I normally go with '16 bit' over the '16 bit linear'
option because the linear file requires a lot of photoshop work to get
it to look right and I haven't seen any benifit.

Thanks,
Patrick
 
Patrick,
For the current version of the LCMS profiler (which you are using), the
"patched" one is the right
one to use. The non-patched one (in the ZIP) has a very fundamental
whitepoint bug, however, the
degree that the bug will affect images will depend on how far off neutral
the raw scans
are to start with. Using the Viewscan software, which does produce
significantly off-neutral
raw scans, the non-patched profiler was unusable IMHO.

Greg.
 
(my comments regarding Viewscan apply to my scanner - not sure about yours.
I have the Nikon LS40)
 
Hi Patrick,

Regarding the nearly white patches in the B row, -try resetting your
scanner by pressing CTRL+SHIFT+I. This has often worked for my scanner
(Minolta Scan Elite II). Might be a good thing to do before you start
scanning a lot of slides.

The beta LCMS is very different from the other LCMS package. You
should be able to produce pretty good profiles for your scanner,
however keep in mind that the film has to be exposed and developed
perfectly... At least you have a much better file to work with after
applying the profile in Photoshop. Some adjustments are often required
but much less than if you didn't have the profile. Remember to NOT
apply any colour space in your scanner driver.

Please tell what you think of your results.

Oistein
 
Hi Patrick,

Regarding the nearly white patches in the B row, -try resetting your
scanner by pressing CTRL+SHIFT+I. This has often worked for my scanner
(Minolta Scan Elite II). Might be a good thing to do before you start
scanning a lot of slides.

The beta LCMS is very different from the other LCMS package. You
should be able to produce pretty good profiles for your scanner,
however keep in mind that the film has to be exposed and developed
perfectly... At least you have a much better file to work with after
applying the profile in Photoshop. Some adjustments are often required
but much less than if you didn't have the profile. Remember to NOT
apply any colour space in your scanner driver.

Please tell what you think of your results.

Oistein
 
Greg,

Are you saying that a raw scan with minimal colour cast is "easier"
for LCMS to handle than a raw scan with e.g. a strong red cast? I
thought the whole point was to let LCMS handle the raw scan (i.e. NOT
colour adjusted in ANY way) so as to produce profiles for colour
correction?!

Oistein
 
Oistein said:
Are you saying that a raw scan with minimal colour cast is "easier"
for LCMS to handle than a raw scan with e.g. a strong red cast?
Click to expand...

Yes, but *only* for the buggy version. The bug is fixed in the "patched"
version. :)
I thought the whole point was to let LCMS handle the raw scan (i.e. NOT
colour adjusted in ANY way) so as to produce profiles for colour
correction?!
Click to expand...

Exactly. If you use the patched version of the profiler, it will do this
properly.
(or the more recent profiler, presumably - I haven't tried it yet, though)

Greg.
 
Oistein,

Yes I did try ctr-shift-i and got better results. The result still
doesn't look exactly like the target slide but it is closer.

What concerns me now is that I'm getting dE maximums of 4 to 6.5. I
get this with diblook and Monaco EZ color as well. Apparently the
human eye can't detect differences of dE=3 to dE=6. I'm getting
worried that I'm right at the edge of the max. How can I lower this? I
tried scanning with different settings but get around the same result.
Does it have to do with the scanner being far off from what the target
is? If so maybe there is something wrong with my scanner. Any thoughts
are welcome.

By the way I'm using a Minolta 5400.

Patrick
 
Patrick said:
Oistein,

Yes I did try ctr-shift-i and got better results. The result still
doesn't look exactly like the target slide but it is closer.

What concerns me now is that I'm getting dE maximums of 4 to 6.5. I
get this with diblook and Monaco EZ color as well. Apparently the
human eye can't detect differences of dE=3 to dE=6. I'm getting
worried that I'm right at the edge of the max. How can I lower this?
Click to expand...
If your profile manges to fit the data points with avg dE <= 1 and max
dE <= 5, then IMO this is a pretty reasonable result for profiling a
consumer scanner with an IT8 target. Since measurements are never
error-free, but noisy, a better fit of the measurement data would rather
fit the noise, and not really increase the accuracy. IMO a good profile
must also retain some amount of smoothness, otherwise you won't be happy
with the results either.

Regards
Gerhard
 
When I used the "old" LCMS, I got an avg dE of about 0.8 - however
when using the beta3 version of LCMS I can't find this value anywhere.
Any idea where I should look? All values returned are much higher (if
any of them reflects the dE), typically in the range 3-13. However,
the target scan looks pretty ok after applying the profile. Any
comments on that?

Oistein
 
Patrick said:
get this with diblook and Monaco EZ color as well. Apparently the
human eye can't detect differences of dE=3 to dE=6. I'm getting
Click to expand...

I don't know what make you think that. I've personally seen examples
of differences <= 1 dE that are just visible, but it depends a lot on
the exact situation, color and expectations. Generally differences
are more obvious near neutral colors, and less obvious near highly
saturated colors.

The CIE94 Delta E weighting (and other similar schemes) reflect
this effect. I've been using the CIE94 weightings in fitting lately,
for matrix/shaper profiles in Argyll to give better visual results.

One thing to keep in mind when throwing around delta E numbers
in regard to profiling "accuracy", is that generally these numbers
reflect the profiles fit to the data points used to create the
profile (self fit). This is not a terribly good guide to real, underlying accuracy.

In some experiments I've done in creating Lut based profiles, I've compared
profiles created with (say) 800 test points, against 8000 reference points
for the same device. If I fiddle some of the magic numbers used in creating
the Lut profile, I can get an almost arbitrary small delta E to the 800 test
points, while measured against the 8000 reference points, the delta E actually
increases. In other words, a better estimate is created of the underlying devices
behaviour, if test values are "averaged" amongst themselves somewhat (reducing
the effects of random measurement errors), even if this is at the cost of increasing
the self fit delta E.

Graeme Gill.
 
When I used the "old" LCMS, I got an avg dE of about 0.8 - however
when using the beta3 version of LCMS I can't find this value anywhere.
Any idea where I should look? All values returned are much higher (if
any of them reflects the dE), typically in the range 3-13. However,
the target scan looks pretty ok after applying the profile. Any
comments on that?
Click to expand...


Agreed. One thing to remember: the error values shown by the profiling
software can not be easily compared, especialy when mean error levels
reach really low values as they do for good profilers. Nor does a low
error value mean there is no major error. Do not forget that the error
values shown by the profiling software are the error values for the
patches of the target also used to create the profile. Error values
for other colors or even other targets will be different...

But in general, seeing a mean error of below 1 can be considered good.
I would start worrying if the mean error gets above dE 1.5-2 as this
usualy means a larger number of patches have higher visible faults...

As far as the max error shown by lcms is concerned: the max error is
often caused by scanner hardware limitations that the profiler can not
fix. Most consumer scanners have problems especialy with highly
saturated greens or very dark (L <3) colors. The max error value
shown thus greatly depends on what limitations the scanner hardware
has and if the target has such colors on it. For instance, for my
Nikon LS 40 lcms does show a max error of 8. But when testing using
1100 other colors not on the target I do find colors with a max error
Click to expand...
up to dE 30.

So, the max error is hardly something the profiler can fix nor is it
likely that the max error shown by the profiler really reflects the
real max error of the complete profile. The main question is, how
does the profiler handle those high error colors. Some profilers cause
strong visible color shifts while others can manage to map colors with
little visible fault. The other question is, how far does the max
error spread into the surrounding. To make it simple: at least with
lcms and most scanners the max error shouldn't be of much concern if
the mean fault is low and the max error is below dE 12. As said the
max error value shown by the profiler hardly reflects the real max
error of the profile...
 
Would anybody comment on the pixel values below?

Just wondering if my profile is ok (used LCMS beta3)- I feel that the
photos often come out with a GREEN cast after applying the profile and
I have to do level corrections in Photoshop to get it good enough for
printing.

IT8 target is PROVIA, ASTIA (Wolf Faust).
Scanner is MINOLTA SCAN ELITE II
Scannersoftware from Minolta, all settings on NONE
The target scan comes out pretty reddish... (as you might see from the
pixel values as well)

Here are some pixel values for the gray patches (before and after
applying profile and converting to AdobeRGB):

GS0
RED 252 GREEN 248 BLUE 249 (withouth profile)
RED 253 GREEN 252 BLUE 252 (profile applied followed by AdobeRGB
conversion)

GS1
231, 224, 230
222, 226, 227

GS2
219, 209, 217
212, 217, 217

GS8
147, 129, 141
141, 145, 142

GS9
135, 117, 128
131, 134, 131

GS10
123, 104, 116
119, 122, 119

GS11
115, 96, 109
111, 114, 113

GS12
103, 83, 95
101, 103, 101

GS16
58, 42, 54
63, 64, 62

Do these values look ok for Provia/Astia??
Thanks for your responses! If Wolf Faust has a comment as well, I'd
really appreciate it. Thanks a lot.

Oistein
 
Would anybody comment on the pixel values below?
Click to expand...

I will wrote down the values as expected to be found on the target from
the IT8 description.
GS0 RED 252 GREEN 248 BLUE 249 (withouth profile) RED 253 GREEN 252
BLUE 252 (profile applied followed by AdobeRGB conversion)
230,227,226

GS1
231, 224, 230
222, 226, 227
201,204,203

GS2
219, 209, 217
212, 217, 217
190,193,192

GS8
147, 129, 141
141, 145, 142
127,129,126

GS9
135, 117, 128
131, 134, 131
118,119,116

GS10
123, 104, 116
119, 122, 119
105,107,104

GS11
115, 96, 109
111, 114, 113
98,100,98

GS12
103, 83, 95
101, 103, 101
88,88,85

GS16
58, 42, 54
63, 64, 62
51,52,50

Do these values look ok for Provia/Astia??
Click to expand...

It seems they are a bit on the bright side, but apart from that they
look reasonable. No color cast at least.

regards
Markus
 
Thanks for your reply,

Here is an example of the GS8 patch (LAB mode values)

Target values: 54 (L) -1 (a) +1 (b)
Values on my scanned IT8 target after profiling and conversion to
AdobeRGB: 60 (L) -2 (a) +1 (b)

Can I interpret anything from this? The values look pretty different
to me.. The values in the other G patches are also quite different
from the LAB values indicated in the target data file that came with
the target.

Thanks!
Oistein
 
Oistein said:
Thanks for your reply,

Here is an example of the GS8 patch (LAB mode values)

Target values: 54 (L) -1 (a) +1 (b)
Values on my scanned IT8 target after profiling and conversion to
AdobeRGB: 60 (L) -2 (a) +1 (b)

Can I interpret anything from this? The values look pretty different
to me.. The values in the other G patches are also quite different
from the LAB values indicated in the target data file that came with
the target.
Click to expand...
Two questions:

a) which rendering intent did you use for the
convertsion from scanner Device-RGB to AdobeRGB

b) how did you convert from AdobeRGB to L*a*b*,
since "60 (L) -2 (a) +1 (b)" does not refer to RGB
values in any RGB color space, but in CIELAB

I can only guess the answer to (a) and (b), but if both
conversions have been done relative colorimetric, then
your values sound pretty reasonable, since the measurements
in the target reference file are illuminant relative, while
applying the profile with relative colorimetric intent gives
you values relative to the media white point.

(And if you've been using "perceptual" intent, then don't
expect anyway to get colorimetrically exact results)

Best Regards
Gerhard
 
Gerhard Fuernkranz said:
Two questions:

a) which rendering intent did you use for the
convertsion from scanner Device-RGB to AdobeRGB
Click to expand...

The rendering intent was set to Perceptual under colour settings in
Photoshop.
b) how did you convert from AdobeRGB to L*a*b*,
since "60 (L) -2 (a) +1 (b)" does not refer to RGB
values in any RGB color space, but in CIELAB
Click to expand...

I just set the Info box to show LAB values (instead of CMYK which is
default).

I can only guess the answer to (a) and (b), but if both
conversions have been done relative colorimetric, then
your values sound pretty reasonable, since the measurements
in the target reference file are illuminant relative, while
applying the profile with relative colorimetric intent gives
you values relative to the media white point.
Click to expand...

So this is not the best way to check if the profile is ok then..?
What is actually the best way to check a profile?
(And if you've been using "perceptual" intent, then don't
expect anyway to get colorimetrically exact results)
Click to expand...

I'll switch to Relative colorimetric - by the way, what is actually
the main difference here? Would I experience any difference on the
final print for instance..?
Best Regards
Gerhard
Click to expand...


Thanks!
Oistein
 
Oistein said:
The rendering intent was set to Perceptual under colour settings in
Photoshop.
Click to expand...
Hello Oistein,

an input profile may contain AtoB tables for either several rendering
intents, or only a single AtoB0 table for only one intent. In the latter
case, the single table can only represent a single intent, and in this
case it is also important, for which intent you have *created* the
scanner profile.
I just set the Info box to show LAB values (instead of CMYK which is
default).
Click to expand...
I'm not sure, how PS performs this conversion, but according to the
values you mentioned, it rather looks like relative colorimetric, or
alternatively it probably chooses the working space's white point as
reference white for the conversion to CIELAB. But I can only guess.
So this is not the best way to check if the profile is ok then..?
What is actually the best way to check a profile?
Click to expand...
Optimally, you should check the profile with colors different from the
patches on the IT8 target you used to create the profile. But I guess,
you don't have another test chart (besides the IT8 target) with
well-known, accurately measured colors available. So your only chance is
check the profile against the IT8 target values. You can use e.g.
qtprofilechecker (comes with the old LCMS profiler) for this job.

If want you do it manually, then you should apply the scanner profile
with ICC-absoulte rendering intent to convert from scanner RGB to XYZ
(or CIELAB) and compare the result of the conversion with the target
reference file. Don't convert scanner RGB to any intermediate RGB
working space before converting to XYZ or CIELAB.
I'll switch to Relative colorimetric - by the way, what is actually
the main difference here? Would I experience any difference on the
final print for instance..?
Click to expand...
The goal of the perceptual intent is a "pleasing result", regardless
whether this result is colorimetrically corret. Thus, besides
colorimetric transformations, a preceptual intent transform usually
contains some "artistic" transormations. So you can never judge a
perceptual intent transformation by comparing measurements, but only by
visual inspection -- either you like the result, or you don't ...

Best Regards,
Gerhard
 
Back
Top