VueScan 7.6.15 now has ICC/IT8 Calibration
Ed Hamrick
This is a free upgrade for registered users (as usual).
You can download it from:
http://www.hamrick.com/vsm.html
What's new in version 7.6.15
* Added support for IT8 calibration and scanner ICC files
* Improved internal ICC profiles (better color accuracy)
There are instructions in the VueScan User's Guide for using
this new capability.
I used this IT8 calibration feature to recalibrate most of
the scanners that VueScan supports. The colors from
VueScan should now look a lot better with most scanners.
Regards,
Ed Hamrick
Jon Bell
Ed Hamrick <use...@hamrick.com> wrote:
>
>I used this IT8 calibration feature to recalibrate most of
>the scanners that VueScan supports. The colors from
>VueScan should now look a lot better with most scanners.
Does this apply to archived raw files from previous versions, or do I have
to scan the film again to take advantage of this?
--
Jon Bell <jtbe...@presby.edu> Presbyterian College
Dept. of Physics and Computer Science Clinton, South Carolina USA
Ed Hamrick
> Does this apply to archived raw files from previous versions, or do I have
> to scan the film again to take advantage of this?
Yes, this applies to archived raw files. You don't need to
scan the film again.
Regards,
Ed Hamrick
Wilfred van der Vegte
> There are instructions in the VueScan User's Guide for using
> this new capability.
Sounds interesting. I checked out this coloraid.de site you mention in
the instructions and I noticed that the targets are film dependent.
To what extent can a calibration based on, for instance, an Ektachrome
target, also be used for scanning, for instance, Fuji Sensia? And is an
ICC/IT8 calibration also used for scanning negatives? I guess that must
be the case because the calibration replaces the 'conventional' VueScan
calibration file - am I right?
> I used this IT8 calibration feature to recalibrate most of
> the scanners that VueScan supports. The colors from
> VueScan should now look a lot better with most scanners.
Is there a list of re-calibrated scanners? Does it include the Minolta
Scan Speed?
--
Wilfred van der Vegte.
Replace 'dot nl' by 'dot net' to reply by e-mail
Ed Hamrick
> To what extent can a calibration based on, for instance, an Ektachrome
> target, also be used for scanning, for instance, Fuji Sensia?
It might work slightly better to calibrate using the same film
type as you're scanning.
> And is an ICC/IT8 calibration also used for scanning negatives?
No, nobody sells ICC/IT8 targets on negative film.
> Is there a list of re-calibrated scanners?
All of the scanners that VueScan supports.
> Does it include the Minolta Scan Speed?
Yes.
Regards,
Ed Hamrick
Wilfred van der Vegte
> "Wilfred van der Vegte" <wi...@gmx.nl> wrote:
>>And is an ICC/IT8 calibration also used for scanning negatives?
> No, nobody sells ICC/IT8 targets on negative film.
Thanks for your quick reply. I'm aware that no one sells negative
targets, but what I meant is that the calibration (even if it is based
on a slide) says something about the color aberrations of an indivisdual
scanner, which could also be used to improve the color accuracy of
negative scans. Couldn't this be used by VueScan?
Ed Hamrick
> Thanks for your quick reply. I'm aware that no one sells negative
> targets, but what I meant is that the calibration (even if it is based
> on a slide) says something about the color aberrations of an indivisdual
> scanner, which could also be used to improve the color accuracy of
> negative scans. Couldn't this be used by VueScan?
VueScan already does this.
Regards,
Ed Hamrick
Bart van der Wolf
"Wilfred van der Vegte" <wi...@gmx.nl> wrote in message
news:b34u2c$4ke$1...@news.tudelft.nl...
> Ed Hamrick wrote:
>
> > There are instructions in the VueScan User's Guide for using
> > this new capability.
>
> Sounds interesting. I checked out this coloraid.de site you mention
in
> the instructions and I noticed that the targets are film dependent.
> To what extent can a calibration based on, for instance, an
Ektachrome
> target, also be used for scanning, for instance, Fuji Sensia?
No.
> And is an ICC/IT8 calibration also used for scanning negatives?
In theory, yes. However, I doubt that the processing variances will
get you better results, and there are more variables involved in
scanning (e.g. heavy film base color and lower contrast/higher scene
dynamic range).
Bart
Wilfred van der Vegte
> "Wilfred van der Vegte" <wi...@gmx.nl> wrote in message
>>To what extent can a calibration based on, for instance, an
>> Ektachrome
>>target, also be used for scanning, for instance, Fuji Sensia?
> No.
That sounds contradictory to Ed's answer, the more because he says
VueScan already uses new target-based corrections for the supported
scanners. I guess Ed used an Ektachrome target (because he has the most
film data on Kodak films), and knowing Ektachrome's properties he was
able to isolate the typical color aberrations of each type of scanner.
If Ed really used Ektachrome for his new corrections, it might even be
preferrable to use an Ektachrome target anyway, if you want more
accurate color corrections for one individual scanner (rather than a
type of scanner).
BTW, what type of target is supplied with scanners where it is included?
My scanner didn't come with a target, but I guess Ektachrome is the
default ... ?
shAf
> Ed Hamrick wrote:
> >
> > No, nobody sells ICC/IT8 targets on negative film.
>
> Thanks for your quick reply. I'm aware that no one sells
negative
> targets, but what I meant is that the calibration (even if it is
based
> on a slide) says something about the color aberrations of an
indivisdual
> scanner, which could also be used to improve the color accuracy
of
> negative scans. Couldn't this be used by VueScan?
Yes ... (Ed will correct me if I'm wrong, hopefully) this is how
it should work. That is, scan a target ... any target with an
accurate reference file, preferably a chrome (i.e., no mask) for
profiling the color gamut of the scanner. All other
characteristics would be corrected via selecting the appropriate
'film type' and accurately measuring & removing the mask.
I have yet to play with Vs 7.6.15 ... but I am a bit leary ...
i.e., my IT8 is 10 years old and I have to wonder how accurate my
reference file is. There is also some question as if the Kodak
IT8 really does approach the true saturation of colors available
from film. So, there inlies the rub ... to use my IT8 strictly
for accurate color, and not as a measure of gamut(?) ... or to
trust Vuescan as is, which already produced accurate color, even
if the gamut was questionable (to the limit gamut can be "seen"
with a monitor).
--
cheerios ... shAf :o)
Avalon Peninsula, Newfoundland
www.micro-investigations.com
Bart van der Wolf
> Bart van der Wolf wrote:
>
> > "Wilfred van der Vegte" <wi...@gmx.nl> wrote in message
>
> >>To what extent can a calibration based on, for instance, an
> >> Ektachrome
> >>target, also be used for scanning, for instance, Fuji Sensia?
>
> > No.
>
> That sounds contradictory to Ed's answer, the more because he says
> VueScan already uses new target-based corrections for the supported
> scanners.
Calibration is based or the interaction of the scanner's lightsource
color, the film's dye set!, and the CCD filter's transmission
characteristics. Changing the dye set from Ektachrome to a Fuji one
will cause differences. How different, depends on the lightsource (and
most also change as they age) and the CCD filters (some scanners
change lightsource color instead of using filters). Also, the scanner
CCD filtration is not identical to the spectrophotometer filtration
when the reference values were determined.
> I guess Ed used an Ektachrome target (because he has the most
> film data on Kodak films), and knowing Ektachrome's properties he
was
> able to isolate the typical color aberrations of each type of
scanner.
> If Ed really used Ektachrome for his new corrections, it might even
be
> preferrable to use an Ektachrome target anyway, if you want more
> accurate color corrections for one individual scanner (rather than a
> type of scanner).
It might, depending on the characteristics of the scanner. Nobody can
make blanket statements for all configurations. And on top of that,
film processing and the unknown storage/handling conditions of film
itself add another variable.
The whole idea is to standardize as much as practical.
> BTW, what type of target is supplied with scanners where it is
included?
> My scanner didn't come with a target, but I guess Ektachrome is the
> default ... ?
Very few do come with a target, and targets fade and get scratched so
they need to be replaced after a while (also the storage conditions of
targets can be standardized). Ektachrome is a kind of a default.
Once you know that the scanner film combination delivers consistent,
predictable colors, then you can start to think about colorbalancing
and tonescaling for your ideal "look".
Bart
UrbanVoyeur
"Ed Hamrick" <use...@hamrick.com> wrote in message
news:b33rj0$r5a$1...@ngspool-d02.news.aol.com...
ThomasH
>
> I just released VueScan 7.6.15. It now contains ICC/IT8 color calibration.
> This is a free upgrade for registered users (as usual).
>
> You can download it from:
>
> http://www.hamrick.com/vsm.html
>
> What's new in version 7.6.15
>
> * Added support for IT8 calibration and scanner ICC files
> * Improved internal ICC profiles (better color accuracy)
>
> There are instructions in the VueScan User's Guide for using
> this new capability.
Something for the very advanced...
A though:
The Color Tab gets very large now. Since the Media Type
is already in the Device tab, maybe the scanner profiles
might be also put to the Device tab, where the input
source is described?
Someone in this thread said the new ICC in 7.6.15 work great,
probably I will have to finish that book by B.Fraser before
I will get that result...
If my assumption is correct that with the
Color|Scanner color space=Built-in
the results should be as previously, than something is wrong.
Colors are really completely different, I will stay with the
7.6.14 at least for the finishing of the scan being in progress
and I will run some tests to verify if I maybe make some
mistakes in setup...
In the "Real World Color Management" Fraser writes about
the "memory colors," such as sky or skin. Some tools have
aids to preserve such colors. For example, PSP has its
"skin colors present" tag in the Light Temperature filter.
Maybe one day Vuescan will get a tag "sky present," what
would simplify the work on the led-gray, greenish or cyan
sky of the White Balance algorithm. The self made ICC
scanner profiling will not resolve these by far larger
discoloration... I am just dreaming.
Thomas
Wilfred van der Vegte
> Calibration is based or the interaction of the scanner's lightsource
> color, the film's dye set!, and the CCD filter's transmission
> characteristics. Changing the dye set from Ektachrome to a Fuji one
> will cause differences.
I guess these are the same differences that we take for granted if we
set the media type to 'image' rather than 'slide' (with a matching
-Kodak-only- slide type), as is the recommended setting ('image', I
mean), and the same differences we take for granted if we are viewing
different types of slides projected by a projector, is't it? It can't be
too bad, then.
> How different, depends on the lightsource (and
> most also change as they age) and the CCD filters (some scanners
> change lightsource color instead of using filters). Also, the scanner
> CCD filtration is not identical to the spectrophotometer filtration
> when the reference values were determined.
(...)
> Once you know that the scanner film combination delivers consistent,
> predictable colors, then you can start to think about colorbalancing
> and tonescaling for your ideal "look".
Yes, there are many considerations, but in my case the colors of slides
are mostly according to my ideal "look" as you call it. Regarding all
these factors that you mention, and that can theoretically disturb the
precious calibration, I think I won't invest in an IT 8 target (or two
of them for the types of slide film I'm using). Scanning negatives is
still the hardest part, anyway.
--
Wilfred van der Vegte
Replace 'dot nl' with dot 'dot net' to reply by e-mail.
Bart van der Wolf
> There are instructions in the VueScan User's Guide for using
> this new capability.
Only the choice of Device color space gives acceptable colors, but it
doesn't embed a profile in the output file. The resulting file looks
suspiciously much like sRGB, and as such would have a limited gamut.
Am I misinterpreting things, of is there something missing? It is also
impossible to install the VueScan created icc file in Windows, in
order for other applications (e.g. Photoshop) to be able and assign it
to the file. If it only is meant for use in VueScan, then I'd prefer
it didn't have the same extension as a real icc file.
Bart
UrbanVoyeur
profiles in my tiff files, with or without calibration.
BTW, now that I have calibration in Vuescan, I have begun to experiment with
BruceRGB. It seems to be closer to the fuji provia I shoot, but is a little
red - as opposed to Adobe RGB, which tend to be a little flat in the reds
and a bit cyan. And yes, I am using an colorimeter calibrated monitor.
"Bart van der Wolf" <bvd...@nospam.nl> wrote in message
news:3e569741$0$49104$e4fe...@news.xs4all.nl...
UrbanVoyeur
"ThomasH" <tho...@coco.net> wrote in message
news:3E5675F2...@oracle.com...
> Ed Hamrick wrote:
> Someone in this thread said the new ICC in 7.6.15 work great,
> probably I will have to finish that book by B.Fraser before
> I will get that result...
>
> If my assumption is correct that with the
>
> Color|Scanner color space=Built-in
>
> the results should be as previously, than something is wrong.
> Colors are really completely different, I will stay with the
> 7.6.14 at least for the finishing of the scan being in progress
> and I will run some tests to verify if I maybe make some
> mistakes in setup...
>
In calibration mode you use the color space of the icc file you produce
during calibration. It is very different form the "built-in" space.
That being said, I do believe that the built-in profiles have been tweaked
by Ed using this new calibration feature.
I would encourage you to get your hands on an IT-8 target, and try the
feature. It makes getting good scans a whole lot easier.
J
UrbanVoyeur
I think you may be surprised at how much closer to calibrated scans are to
the original look of your slides. Yes, as with any calibration (scanner,
monitor, printer) it is important to re-calibrate regularly. However, IMO,
calibration will have a significant positive impact on the quality of your
scans and the time it takes to tweak them.
It has been my most sought after feature in vuescan, and so far, it works
great.
now all I need is to convince Wolf Faust to produce 2 new targets - 1 for
velvia 50 & 100, and 1 for the new Astia.
Bill Hilton
>BTW, now that I have calibration in Vuescan, I have begun to experiment with
>BruceRGB.
Have you looked at Joe Holmes' Ektaspace 5? It's supposed to mimic the gamut
of slide film very closely, better than BruceRGB, or so one hears. (Just
curious as to how it stacks up)
Greg
target.
I have the same problem as you - I can't "install" the ICC profile which
Vuescan generates.
So far, I'm not confident that I am getting the most out of Vuescan with ICC
profiling, or, that it's even working properly. For example, if I save the
IT8 target scan from Vuescan (using the just-created ICC profile for the
scanner profile), telling Vuescan to use a file space of Ektaspace, the
gamut doesn't appear to be large as the gamut I obtain when using Nikon
Scan. (I have profiled my LS40/Nikon-Scan combination as well). I am making
this judgement by using Photoshop 7's "gamut warning" function, using
Relative Colorimetric for the proofing method, and sRGB for the proof
profile. Looking at Vuescan's scan in this way, just a few of the darker
yellow bars appear to be out of gamut. Looking at the result from Nikon
Scan, many, many more of the swatches are out of gamut. I thought that
perhaps the restriction of gamut might have something to do with the fact
that Vuescan is doing a conversion to Ektaspace, but I have also done a
conversion of my Nikon Scan from my custom profile to Ektaspace, and
repeated the gamut warning view, and it doesn't change much - still appears
that the Nikon Scan result has a wider gamut.
I assume that I should be setting Color Balance to "none" in Vuescan, when
using ICC, yes?
Just by the way, I notice that the "raw" scan from Nikon Scan looks quite a
lot more neutral and also a bit brighter than the raw scan from Vuescan. (I
am doing a non slope-limited gamma adjustment to 2.2 of the Vuescan raw scan
before viewing it). I don't think there is any way to *totally* disable
Nikon Scan's colour management, and this might be why it looks "better". For
example, there appear to be many ICC files in Program Files\Common
Files\Nikon\Profiles, and these *cannot* be deleted for Nikon Scan to work,
even if colour management is disabled in Nikon Scan.
Greg.
"Bart van der Wolf" <bvd...@nospam.nl> wrote in message
news:3e569741$0$49104$e4fe...@news.xs4all.nl...
>
ThomasH
Ok, thanks for the advice! I will give it a shot.
Thomas.
Gary Hundt
> "Wilfred van der Vegte" <wi...@gmx.nl> wrote:
> > To what extent can a calibration based on, for instance, an Ektachrome
> > target, also be used for scanning, for instance, Fuji Sensia?
>
> It might work slightly better to calibrate using the same film
> type as you're scanning.
>
> > And is an ICC/IT8 calibration also used for scanning negatives?
>
> No, nobody sells ICC/IT8 targets on negative film.
What about photographing (with negative film) a reflective target such
as Wolf Faust's "C1" item described as a "DIN A4 sized reflective
camera target on non-glossy Kodak Prof. paper and polystyrol backing.
IT 8.7 layout"?
Or what about a macro "copy" to negative film of an IT8 slide?
I suppose there would be all kinds of variables I'd need to control to
make this viable, most importantly the color temp of the lighting (I'd
think a "cloudy bright" outdoor exposure would be ideal, but obviously
hard to reproduce, so my trusty strobe would probably be my "standard"
source.)
And I suppose I'd want to be prepared to re-calibrate each roll to
account for processing variations... so in practice my "advanced
workflow" routine would be expanded to include IT8 calibration.
I recall earlier discussions here on using a grey card (on each roll
of film) to aid in color balancing. Well, it seems that this IT8
calibration process could serve that purpose at least.
Gary
BTW, Wow! This is a real milestone! Adding this feature to Vuescan has
been a wish of many for a long while. I'm surprised you didn't up the
version number to 8.0!
Wolf Faust
wrote:
> I have yet to play with Vs 7.6.15 ... but I am a bit leary ...
>i.e., my IT8 is 10 years old and I have to wonder how accurate my
>reference file is. There is also some question as if the Kodak
>IT8 really does approach the true saturation of colors available
>from film. So, there inlies the rub ... to use my IT8 strictly
>for accurate color, and not as a measure of gamut(?) ... or to
>trust Vuescan as is, which already produced accurate color, even
>if the gamut was questionable (to the limit gamut can be "seen"
>with a monitor).
Never change a running system ? ;-)
10 years if not stored in the fridge is surely a long time. The Kodak
targets does not show the real chrominance limits of the film. There
are technical limitations during production of the targets that are
not easy to deal with. I am not sure why Kodak is far from reaching
the limits possible with their films, but I guess the devices
available by the time Kodak developed it's target production were not
able to produce bigger gamuts. The production method was not updated
since than. In order to see what is possible with Ektachrome film,
check out the reference files of one of my targets on
http://www.targets.coloraid.de . Compare the luminance values of the
GS0/GS23/A16/L16 patches with those of Kodak. Also check the
chrominance values of the patches in column 4,8,12. You will find that
the coloraid targets do have the biggest color gamut and are pretty
close to the real limits of the film.
--
Wolf Faust EMail: wfa...@wfaust.de
Tel: ++49-69-5486556 WWW: www.coloraid.de
Mobile: ++49-179-6924769 Fax: ++49-69-95409598
Wolf Faust
<sulliva...@bigpond.net.au> wrote:
If I do understand the method you are using correctly, you are
scanning using Nikonscan wich than embeds one of it's generic profile
wich is most likely far from being correct (my LS40 experience). This
might mean that certain colors are not correctly calculated by
PhotoShop and might end up being displayed as out of gamut simply
because the color values are wrong.
But you can find out:
Take a scan of the Provia target and do as you described. But this
time do convert the image in PhotoShop to Lab space. Now use the info
window in order to display the Lab color values of the various target
patches and compare with the correct values of the reference. Good IT
8 scanner profiling software should have a mean/max error below dE
1/6 wich also shows that you will hardly beat the color quality of an
IT 8 calibration by adjusting colors manualy. If PhotoShop does not
display color values close to those defined in the reference file of
your target something went wrong somewhere and the gamut indicator can
not work correctly.
I am not familiar with VueScan. Can you save the scanner data without
color correction but with an embedded profile (set file color space to
Device RGB?) ? Or does VueScan always convert the RGB colors of the
scanner using an input->output profile? Both methods should be
possible. With embedded scanner profile you should also be able to see
the correct Lab values in PhotoShop. If not, again, something went
wrong somewhere...
Note: the target patches in column 4, 8, 12 of the Provia target have
maxiimum chrominance and are on the gamut border of the film. Most of
them are most likely outside of the sRGB gamut. Because of the way the
IT 8 target was produced, you will hardly reach those gamut limits in
practice using normal equipment. So there is some room to limit the
input gamut before you will cut off colors in a real slide.
>I assume that I should be setting Color Balance to "none" in Vuescan, when
>using ICC, yes?
For what you are doing yes. But this option is for changing the colors
of the scene... while the ICC calibration is mainly for helping the
scanner correctly scan film as is. I see no reason otherwise why to
not use the Color Balance feature in combination with ICC profiles.
Just in your case as you want unaltered color values...
UrbanVoyeur
with fuji films as it does with kodak.
"Bill Hilton" <bhilt...@aol.comedy> wrote in message
news:20030221211919...@mb-bk.aol.com...
Greg
For some reason, I always find that the IT8 swatch H8, which is a sort of
turquoise hue (greeny blue, anyway), just never looks right on a CRT
monitor. Recently, I've finally made the discovery that this colour is
simply outside the gamut of sRGB, which I think explains why the monitor
can't reproduce it very well. My Epson 1270 can reproduce this colour
entirely satisfactorily, but *only* when I use "no color adjustments" in the
driver, and do the appropriate profile conversion before printing. Print
this colour after converting to sRGB, and it comes out looking exactly like
it does on screen.
Now, if I bring the IT8 target scan into Photoshop, after scanning it in
Vuescan into Ektaspace (using the custom profile for the scanner, as
generated earlier by Vuescan), and convert the image into lab space, the L
a b values I get are:
51 -19 -32. Looking at the reference file for the target
(39.91, -33.92, -38.25), this represents a deltaE of 19.6, which is not very
good (is it? I don't really know how bad it is, but it *seems* bad, from the
small amount of research I have done). The result from Nikon Scan seems a
*lot* better: 42 -35 -38, which is a deltaE of just 2.4!!! (if I convert
from the custom profile for Nikon Scan to Ektaspace, using relative
colorimetric, there is only a *very* slight shift in the colour, and the
deltaE actually reduces slightly)
If I now convert the Nikon Scan scan to sRGB, using relative colorimetric,
the resulting colour is: -45 -15 -33, and the deltaE soars to 20.3, which is
about the same as the result from Vuescan.
Ed: is it possible that you are restricting the gamut somehow?
Greg.
"Greg" <sulliva...@bigpond.net.au> wrote in message
news:iaC5a.8216$VH5....@news-server.bigpond.net.au...
Greg
news:3e572788...@news.t-online.de...
> If I do understand the method you are using correctly, you are
> scanning using Nikonscan wich than embeds one of it's generic profile
> wich is most likely far from being correct (my LS40 experience). This
> might mean that certain colors are not correctly calculated by
> PhotoShop and might end up being displayed as out of gamut simply
No, that is not correct. I am configuring Nikon Scan to use "Scanner
Profile", which effectively turns off Nikon's colour management. (to the
largest extent possible, anyway). I used this setting when *creating* my own
custom profile for Nikon Scan. I used the Little CMS profiling utilities to
create my profile. I feel that my profile is working very well, and Little
CMS reports a very low maximum deltaE for my profile.
Note that the reason I use "Scanner Profile", and not simply "no colour
management" is simply so Nikon Scan will reliably scan with a gamma of 2.2
when using 12-bits per pixel. If I use "no colour management", it sometimes
seems to use a gamma of 1, but only in 12bpp mode. Strange. It is my
understanding that "no colour management" is identical to "Scanner Profile",
except that when one uses "Scanner Profile", the various adjustments in the
Tools palette are still able to be used. Naturally, I set all those to
neutral when I created my custom profile.
> because the color values are wrong.
>
> But you can find out:
Wow. I had just posted my results, when I came across this post from yours,
which advises me to do what I have just done. :) :) :)
>
> Take a scan of the Provia target and do as you described. But this
> time do convert the image in PhotoShop to Lab space. Now use the info
> window in order to display the Lab color values of the various target
> patches and compare with the correct values of the reference. Good IT
> 8 scanner profiling software should have a mean/max error below dE
> 1/6 wich also shows that you will hardly beat the color quality of an
> IT 8 calibration by adjusting colors manualy. If PhotoShop does not
> display color values close to those defined in the reference file of
> your target something went wrong somewhere and the gamut indicator can
> not work correctly.
See the other post I have just posted for my results for swatch H8. I think
something might be wrong with Vuescan (or the way I am using it) at the
moment.
>
> I am not familiar with VueScan. Can you save the scanner data without
> color correction but with an embedded profile (set file color space to
> Device RGB?) ? Or does VueScan always convert the RGB colors of the
> scanner using an input->output profile? Both methods should be
> possible. With embedded scanner profile you should also be able to see
> the correct Lab values in PhotoShop. If not, again, something went
> wrong somewhere...
Vuescan does not seem to be able to embed the device profile, either the
inbuilt one, or the custom made one. Since the custom profile Vuescan
generates can't be installed (in Windows XP, at least), this suggests that
it might not be completely standard, so I wouldn't really want to embed it
anyway, without understanding more about the profile.
>
> Note: the target patches in column 4, 8, 12 of the Provia target have
> maxiimum chrominance and are on the gamut border of the film. Most of
> them are most likely outside of the sRGB gamut. Because of the way the
> IT 8 target was produced, you will hardly reach those gamut limits in
> practice using normal equipment. So there is some room to limit the
> input gamut before you will cut off colors in a real slide.
>
> >I assume that I should be setting Color Balance to "none" in Vuescan,
when
> >using ICC, yes?
>
> For what you are doing yes. But this option is for changing the colors
> of the scene... while the ICC calibration is mainly for helping the
> scanner correctly scan film as is. I see no reason otherwise why to
> not use the Color Balance feature in combination with ICC profiles.
> Just in your case as you want unaltered color values...
Thanks for the info.
Greg.
Ed Hamrick
> BTW, Wow! This is a real milestone! Adding this feature to Vuescan has
> been a wish of many for a long while. I'm surprised you didn't up the
> version number to 8.0!
I considered using 7.7 or 8.0, but I have some more ideas for
improving VueScan before doing this.
Regards,
Ed Hamrick
Ed Hamrick
> Now, if I bring the IT8 target scan into Photoshop, after scanning it in
> Vuescan into Ektaspace (using the custom profile for the scanner, as
> generated earlier by Vuescan), and convert the image into lab space, the
L
> a b values I get are:
> 51 -19 -32. Looking at the reference file for the target
> (39.91, -33.92, -38.25), this represents a deltaE of 19.6, which is not
very
> good (is it? I don't really know how bad it is, but it *seems* bad, from
the
> small amount of research I have done). The result from Nikon Scan seems a
> *lot* better: 42 -35 -38, which is a deltaE of just 2.4!!! (if I convert
> from the custom profile for Nikon Scan to Ektaspace, using relative
> colorimetric, there is only a *very* slight shift in the colour, and the
> deltaE actually reduces slightly)
One thing you could try to diagnose this is to do this same analysis
of the raw scan file using the ICC profile produced by VueScan.
This will give an indication of whether the problem is with the
matrix produced by VueScan or whether the problem is later in the
processing steps within VueScan.
Regards,
Ed Hamrick
Greg
>
> One thing you could try to diagnose this is to do this same analysis
> of the raw scan file using the ICC profile produced by VueScan.
Ok, I've done that now. (the reason I hadn't done that yet is that Windows
XP won't let me "install" the profile which Vuescan creates. However, I've
now found that just plonking it in the same folder as all the other profiles
allows Photoshop to find it, and use it - Photoshop does not report any
errors). Small suggestion - allow the name of the profile to be edited, OR,
don't give it a name at all - I think when there is no name in the ICC file,
applications will then simply report the file name. The reason for this is
that it shows up as simply "Scanner Profile" at the moment, which isn't very
descriptive.
The L*a*b values for patch H8, when using the raw scan, in conjunction with
the the Vuescan created profile, are:
59, -25, -37. This represents a dE of about 22.
Greg.
Greg
Just a bit more info on how I am scanning in Nikon Scan. When Nikon Scan's
colour management is set to "Scanner Profile", Nikon Scan doesn't actually
embed any profile in the resulting file. This mode is supposed to disable
Nikon's colour management (as described in my previous post). However, Nikon
supplies an ICC profile for the scanner, which can be used with these raw
scans.
I am not using the supplied Nikon profile - I am using a profile I made
myself. I manually tag the scan with the custom profile, in Photoshop, when
I open the scan.
Note that when I use the TWAIN interface of Nikon Scan, regardless of
whether I have Nikon Scan's colour management off or on, I always have to
manually tag the imported image with a profile manually - the profile which
is selected in Nikon Scan is not sent to Photoshop. I'm not sure whether
this behaviour is correct or not - I sort of expected the profile
information to be somehow sent across as well.
Greg.
"Greg" <sulliva...@bigpond.net.au> wrote in message
news:T%G5a.9152$VH5....@news-server.bigpond.net.au...
Ed Hamrick
> The L*a*b values for patch H8, when using the raw scan, in conjunction
with
> the the Vuescan created profile, are:
> 59, -25, -37. This represents a dE of about 22.
Hmm, the matrix profile VueScan creates minimizes the mean square
error in XYZ space. I'll look into minimizing the error in L*a*b
space, since minimizing perceptual error is the main goal of profiling.
I'm also weighting all the patches equally, but should probably
weight the grey patches less.
I should have something to test early next week.
Regards,
Ed Hamrick
Greg
profiling.
For what it's worth, I am able to create a good quality profile from the raw
scan from Vuescan, using the Little CMS profiler. But, since LCMS is only
able to use 8-bits per channel scans, I had to gamma adjust the raw file
first, because the profile of the linear scan actually wasn't that good.
(high max dE: 23 or thereabouts). The max dE when using the gamma adjusted
(I adjusted it to 2.2) scan is only 1.92, which is very good. I did this
test just to satisfy myself that Vuescan's raw scan is capturing the target
well enough.
I have not yet attempted to use this LCMS generated profile in Vuescan.
Greg.
"Ed Hamrick" <use...@hamrick.com> wrote in message
news:b37qaa$o3u$1...@ngspool-d02.news.aol.com...
UrbanVoyeur
Perhaps consider weighing the blue and reds slightly more? Most scanners
seem to exhibit the most problems with red blooming and blue
undersaturation...
Could you please add as "done" notification to the calibration. It took me a
while to realize it was finished...
Also: I noticed that not all the patches in the chart were outlined. Is that
because you don't need the color data on the right of the chart?
"Ed Hamrick" <use...@hamrick.com> wrote in message
news:b37qaa$o3u$1...@ngspool-d02.news.aol.com...
Ed Hamrick
> I have not yet attempted to use this LCMS generated profile in Vuescan.
Do you know if this is a matrix profile? If the file size is much
more than a few KBytes, then it may not work with VueScan.
Regards,
Ed Hamrick
Ed Hamrick
> Also: I noticed that not all the patches in the chart were outlined. Is
that
> because you don't need the color data on the right of the chart?
Different targets have different patches on the right. I
didn't want to have to read the IT8 description file before
drawing the outlines. VueScan uses every patch that's in the
IT8 description file to do the calibration.
Regards,
Ed Hamrick
UrbanVoyeur
Is that the difference between the long and short IT-8 files that Wolf
distributes with his targets?
"Ed Hamrick" <use...@hamrick.com> wrote in message
news:b38b5n$3bt$1...@ngspool-d02.news.aol.com...
Ed Hamrick
> Is that the difference between the long and short IT-8 files that Wolf
> distributes with his targets?
They're just text files - you can look at them with a text editor
and see the differences.
Regards,
Ed Hamrick
Greg
I'm pretty sure it's a lookup table profile - it's about 240Kbytes in size.
The other issue is that even if were the right type, it would not work,
because the profile is for a gamma 2.2 raw scan - I had forgotten about
that.
Greg.
"Ed Hamrick" <use...@hamrick.com> wrote in message
news:b38av3$2qm$1...@ngspool-d02.news.aol.com...
Ed Hamrick
> I'm pretty sure it's a lookup table profile - it's about 240Kbytes in
size.
That's probably why it works well. A lookup table profile will
work well when there's sensor non-linearity, but a matrix profile
won't. It looks like the Nikon scanners aren't completely
linear through the whole intensity range.
I'm going to experiment with correcting the sensor nonlinearity
in the profiles.
> The other issue is that even if were the right type, it would not work,
> because the profile is for a gamma 2.2 raw scan - I had forgotten about
> that.
These still have the rXYZ, gXYZ and bXYZ tags, so they should still
work with VueScan raw files. VueScan ignores the gamma setting the
profile, so it should work. However, VueScan won't take full advantage
of the profile, since it doesn't use the other parts of the profile.
Also, I modified VueScan to produce the 3x3 matrix in a way that
minimizes the root mean square of dE. In this case, the RMS dE
is 9.1. I suspect that I'll need to correct for sensor linearity
to reduce this further.
Regards,
Ed Hamrick
Greg
> Calibration is based or the interaction of the scanner's lightsource
> color, the film's dye set!, and the CCD filter's transmission
> characteristics. Changing the dye set from Ektachrome to a Fuji one
> will cause differences. How different, depends on the lightsource (and
> most also change as they age) and the CCD filters (some scanners
> change lightsource color instead of using filters). Also, the scanner
> CCD filtration is not identical to the spectrophotometer filtration
> when the reference values were determined.
FYI, I have the Ektachrome and Provia targets from Wolf Faust, so I decided
to do an experiment to see how well my Provia profile worked when scanning
the Ektachrome target. My scanner is a Coolscan IV, and I am using Nikon
Scan with it's colour management disabled. The max dE increased from 2.24
to 7.19, and the average dE goes from 0.41 to 2.78. The max dE occurs on a
very dark green patch (J22), and on screen, it's very hard to notice the
difference in colour. If I ignore just this one patch, the next lowest dE is
about 3.9.
In this *particular* configuration, I personally feel that Ed's earlier
comment about the difference being "slight" is correct.
That said, being a purist and wanting to get every last bit of accuracy, I
am still going to buy Wolf's Velvia target when it is available. ;^)
Greg.
Wolf Faust
<nos...@nospam.urbanvoyeur.com> wrote:
>Is that the difference between the long and short IT-8 files that Wolf
>distributes with his targets?
No. The "long" version does include statistical data not used by
profilers. Some profilers do crash on loading the long version (bad
programmers) wich is why I do include a short version without the
statistical data. As far as the profiling is concerned, the long and
the short version should behave exactly the same.
As Ed points out, you can load the IT 8 reference file into a text
editor. The statistical data gives you an idea on the fault each
target has compared to the reference file and allows you to compare
the quality of targets between manufacturers. I do not know any
program that makes actual use of this data. The IT 8 standard does
require manufacturers to provide this data.
Wolf Faust
<sulliva...@bigpond.net.au> wrote:
>Note that when I use the TWAIN interface of Nikon Scan, regardless of
>whether I have Nikon Scan's colour management off or on, I always have to
>manually tag the imported image with a profile manually - the profile which
>is selected in Nikon Scan is not sent to Photoshop. I'm not sure whether
>this behaviour is correct or not - I sort of expected the profile
>information to be somehow sent across as well.
Thanks for the info. I actually only used my Nikon scanner once so
far.
Back to your main problem with the gamut. As it becomes more clear
from the postings here that the profile generated by VueScan uses a
3x3 matrix, your gamut problem does make a lot of sense and the high
fault of dE 19 you measured is very likely real. Maybe someone here
can describe a nice workflow using VueScan with lcms or iphoto 3D LUT
profiles.
Wolf Faust
wrote:
>I'm going to experiment with correcting the sensor nonlinearity
>in the profiles.
If you need help/source/info to get things solved quicker... let me
know...
>These still have the rXYZ, gXYZ and bXYZ tags, so they should still
>work with VueScan raw files. VueScan ignores the gamma setting the
>profile, so it should work. However, VueScan won't take full advantage
>of the profile, since it doesn't use the other parts of the profile.
The ICC standard only allows either Lab (3D LUT) or XYZ (matrix)
profiles but not both. If I remember correctly, Marti did add the
rXYZ, gXYZ and bXYZ tags because some applications did not support the
3D LUTs and required the tags despite the fact that the profile was
for Lab. So basicly something for "illegal" applications I wouldn't
recommend repeating ;-)
>Also, I modified VueScan to produce the 3x3 matrix in a way that
>minimizes the root mean square of dE. In this case, the RMS dE
>is 9.1. I suspect that I'll need to correct for sensor linearity
>to reduce this further.
Yes, but you won't get far with a 3x3 matrix. I think I do have some
articles describing how to improve the quality by not using mean
square dE for building the 3x3 matrix and use a more visual appealing
solution. Also, the result can be greatly improved by proper use of
the 1D Luts in the profile. But no doubt, this will not get you far
compared to the quality of normal 3D LUT profiles found nowadays. Even
a simple calculation based on an order 2-3 polynom regression will
give you much better results (and still most likely fail on colors
close to the gamut border).
Wolf Faust
>What about photographing (with negative film) a reflective target such
>as Wolf Faust's "C1" item described as a
This is not as easy as it might seem. You need a basic knowledge on
how negative film and color does work. Definitly nothing you can
currently hand out to normal users expecting reliable results in all
situations.
>Or what about a macro "copy" to negative film of an IT8 slide?
You could make color targets on negative film and yes, this would
solve many problems. But as you note, this doesn't solve all the
problems. Some variables you need to correct for proper negative
scanning:
1. intensity of scene illuminant
2. color of scene illuminant
3. scene subject matter
4. camera lens transmission color
5. lens aperture, exposure time and film speed
6. film color balance
7. film latent-image keeping properties
8. film processing
9. scanner (spectral sensitivity,...)
So a film target will not solve all issues above. But it should get
you pretty far. And than you might solve the many remaining issues by
certain automatic controls or by using a small, cheap and simple
target. But this has to be developed... and I guess requires 4 months
work for a first version wich adds up to a nice bill and it has to be
done inside a scanner driver to be easily usable. Wich means Ed has to
do it and we all know a day only has 24 hours...
>I recall earlier discussions here on using a grey card (on each roll
>of film) to aid in color balancing.
In the 1940s Kodak did expose a small patch to an unexposed area of
the film using uniform light before processing. After processing the
red, green and blue density were measured and the values were punched
as small holes along the edge as guide for proper exposure of the film
to paper. But this was soon replaced by more reliable methods...
Wolf Faust
<sulliva...@bigpond.net.au> wrote:
>FYI, I have the Ektachrome and Provia targets from Wolf Faust, so I decided
>to do an experiment to see how well my Provia profile worked when scanning
>the Ektachrome target. My scanner is a Coolscan IV, and I am using Nikon
>Scan with it's colour management disabled. The max dE increased from 2.24
>to 7.19, and the average dE goes from 0.41 to 2.78. The max dE occurs on a
>very dark green patch (J22), and on screen, it's very hard to notice the
>difference in colour. If I ignore just this one patch, the next lowest dE is
>about 3.9.
Nice info. Thanks. The difference also depends on the scanner and very
important: profiling software. So the above values do not apply to all
setups and might be much higher with a different scanner. But your
values do make sense and are on the lower end of other studies. A
small but visible color fault even for the untrained eye is what you
can expect and this is also what several users did report (usualy
users switch from Ektachrome to Provia target and suddenly the red
cast in Provia films is gone :-)
But be carefull, the Provia film has a higher dmin and thus you might
cut-off highlights when using the Provia profile for scanning
Ektachrome.
Greg
from the raw Vuescan scan, against CIE(Y) from the target reference. Since
the raw scan has a gamma of 1, this should, I think, result in a linear
response.
The result is here: http://users.bigpond.net.au/sullivag/ls40_linearity.png
I see a few kinks, which I didn't expect. I expected a sort of tapering off
at one or both extremeties, or perhaps a smooth curve, but not little kinks
like that. Still, the response does look quite linear.
To obtain the pixel values, I did a radius 5 gaussian blur of the patches in
Photoshop, and then used Picture Window to obtain the full 16-bit pixel
values of the patches. When I did the "picking", I did all patches for one
channel, and then started over and did the next channel. This means, IMHO,
that the kinks aren't simply due to me clicking on non representative parts
of the patches.
FWIW.
Greg.
"Ed Hamrick" <use...@hamrick.com> wrote in message
news:b38ute$hfb$1...@ngspool-d02.news.aol.com...
Greg
linearity test. Target is Wolf Faust's Provia/Astia.
Greg.
"Greg" <sulliva...@bigpond.net.au> wrote in message
news:eE36a.20137$VH5....@news-server.bigpond.net.au...
Ed Hamrick
> Back to your main problem with the gamut. As it becomes more clear
> from the postings here that the profile generated by VueScan uses a
> 3x3 matrix, your gamut problem does make a lot of sense and the high
> fault of dE 19 you measured is very likely real. Maybe someone here
> can describe a nice workflow using VueScan with lcms or iphoto 3D LUT
> profiles.
There were several reasons for the high dE (on one patch).
1) Nonlinear CCD and A/D converter
2) VueScan was minimizing mean square error in XYZ space
I've fixed both of these things in VueScan 7.6.16, and will
release it in the next hour or so.
VueScan 7.6.16 now produces a mean dE of approximately 3.1
and a worst-case dE of about 10 on a few patches.
There are a variety of reasons that I want to continue using
a matrix ICC profile, both because of speed and because of
the way I do correction of color negative film.
I get e-mails daily from people asking me to speed up VueScan,
so I'm quite sensitive about adding features (like 3D LUT
profiles) that could significantly slow down VueScan. I
also think the current color accuracy is excellent, and is
mostly limited by sensor nonlinearity.
People are also generally quite happy with the colors that
VueScan gets with color negative film, and using a matrix
ICC profile allows me to continue doing this.
Regards,
Ed Hamrick
Ed Hamrick
> Yes, but you won't get far with a 3x3 matrix.
It's working quite nicely now.
> I think I do have some
> articles describing how to improve the quality by not using mean
> square dE for building the 3x3 matrix and use a more visual appealing
> solution.
Could you let me know what this technique was? I'd like to
experiment with this.
> Also, the result can be greatly improved by proper use of
> the 1D Luts in the profile.
Yes, this was very important.
> But no doubt, this will not get you far
> compared to the quality of normal 3D LUT profiles found nowadays.
The matrix profiles are usable with the first part of
color negative processing, since they allow fixing sensor
nonlinearities and allow transforming the colors to a
canonical color space before doing orange mask removal.
Regards,
Ed Hamrick
Bart van der Wolf
SNIP
> In this *particular* configuration, I personally feel that Ed's
earlier
> comment about the difference being "slight" is correct.
Thanks for that feedback. But as said, the difference depends on the
particular configuration.
> That said, being a purist and wanting to get every last bit of
accuracy, I
> am still going to buy Wolf's Velvia target when it is available. ;^)
That's the spirit!
Bart
Ed Hamrick
> I see a few kinks, which I didn't expect. I expected a sort of tapering
off
> at one or both extremeties, or perhaps a smooth curve, but not little
kinks
> like that. Still, the response does look quite linear.
The Nikon scanners are quite linear, generally requiring only
a power factor correction of between 0.97 to 1.03.
The PhotoSmart scanner I tested needed a more extreme power
factor correction in the green channel.
I think the kinks are due to internal reflections. You can see
this sometimes when a bright patch is in the same scan line as
a darker area - sometimes the patch is a higher intensity than
it should be (at least with a few scanners I've looked at closely).
Regards,
Ed Hamrick
Bart van der Wolf
> I've calibrated my LS40 now, using Vuescan, and Wolf Faust's
Provia/Astia
> target.
> I have the same problem as you - I can't "install" the ICC profile
which
> Vuescan generates.
SNIP
> I assume that I should be setting Color Balance to "none" in
Vuescan, when
> using ICC, yes?
It doesn't say in the instructions, and frankly I don't think it
matters for the profiling itself, because the profiling is done based
on the raw data before colorbalancing. However, the (gamma corrected)
output is of course influenced by whatever color balance you choose.
And I have a problem there.... I'll try again with a new download.
Bart
ThomasH
>
> On Sat, 22 Feb 2003 03:31:26 GMT, "Greg"
> <sulliva...@bigpond.net.au> wrote:
>
[...]
> Take a scan of the Provia target and do as you described. But this
> time do convert the image in PhotoShop to Lab space. Now use the info
> window in order to display the Lab color values of the various target
> patches and compare with the correct values of the reference. Good
Wolf,
what is the current status of the Agfa targets? You wrote on
your webpage that they would be available at the end of 2002.
Will be target A3 be applicable on CT100 and CT200 Precisa as well?
Thomas
Greg
Note that I made a silly mistake in my test - I used the Ektachrome target
info, whereas the scan was from the Provia target. So for now please don't
make too much of the kinks.
Greg.
"Ed Hamrick" <use...@hamrick.com> wrote in message
news:b3am1j$pd0$1...@ngspool-d02.news.aol.com...
Greg
data into my spreadsheet and had a look - the result looks better - hardly
any kinks at all. :) (funny that!) Will upload the new screen grab later,
FWIW.
Greg
Gerhard Fuernkranz
Ed Hamrick schrieb:
> "Wolf Faust" <ju...@coloraid.de> wrote:
...
>>Also, the result can be greatly improved by proper use of
>>the 1D Luts in the profile.
>
> Yes, this was very important.
Yes, I also fully agree with Wolf, that you won't get too far with a matrix
only profile, except for scannerns which a priori have a very linear response,
even without correction.
E.g. with my Epson 1240U Photo flatbed scanner I experienced gamma differences
of mor than 0.2 between the R, G, and B channel. With a matrix correction
only, I'll never get reasonable results with this scanner. But with a simple
gamma+matrix profile I can achive an average dE of about 2..3, which isn't too
bad, compared to the uncorrected scan. With a shaper+matix profile (1D LUT +
matix) with higher order curves I get an average dE of below 2 and with a 3D
LUT < 1.5 (depending on the resolution of the LUT).
Due to the fact, that scanners are RGB devices, where each color channel has
its non-linearity, but the channels are relatively independent, a 1D shaper +
matrix usually results in relatively good results for scanners (compared to
the uncorrected data), but of course a 3D LUT can still improve the profile,
particularly the max. dE of the profile.
BTW: Ed, I do not agree that 3D LUTs are necessarily slow. Have you ever tried
the highly optimized IMDI routines from Graeme Gill's Argyll CMS
(http://web.access.net.au/argyll/argyllcms.html)? Once the tables are set up,
the transformations are really fast and not much slower than matrix/shaper
transformations. I was really surprised about the speed of argyll's cctiff
command :-)
>>But no doubt, this will not get you far
>>compared to the quality of normal 3D LUT profiles found nowadays.
>
> The matrix profiles are usable with the first part of
> color negative processing, since they allow fixing sensor
> nonlinearities and allow transforming the colors to a
> canonical color space before doing orange mask removal.
I'm wondering whether it would be possible to create a 2nd "negative profile"
for negative scanning, using the same (positive) IT8 target slide, but with
the scanner set to "negative" mode, just in order - as you said - to correct
the scanner's impacts and to convert the scanned negatives to a canonical
color space prior to orange mask processing. I guess this could work, if the
density range of the IT8 slide is larger than the density range of negative
film, however I don't know if this is generally granted? Wolf, do you know?
BTW: Basically, a given profile is only accurate for one particular set of
scanner settings, i.e. those setting which were in effect at the time, when
the profile was created. For different scanner settings, e.g. different
exposure, positive vs. negative scannning, etc. actually a different profile
should to be created and used. Therefore I think a profile created with the
"transparency/positive" setting of a scanner is not representative for
"transparency/negative" scanning anyway.
Best Regards,
Gerhard
Ed Hamrick
> Yes, I also fully agree with Wolf, that you won't get too far with a
matrix
> only profile, except for scannerns which a priori have a very linear
response,
> even without correction.
That's why I modified VueScan 7.6.16 to also do a per-color
gamma (i.e. power) correction. This worked pretty well.
You can download VueScan 7.6.16 from:
http://www.hamrick.com/files/vuesca76.sit
> E.g. with my Epson 1240U Photo flatbed scanner I experienced gamma
differences
> of mor than 0.2 between the R, G, and B channel.
The Nikon scanners are more linear than this, but other scanners
are less linear. This is why anyone serious about getting good
color will use an IT8 calibration target.
> With a matrix correction
> only, I'll never get reasonable results with this scanner. But with a
simple
> gamma+matrix profile I can achive an average dE of about 2..3, which isn't
too
> bad, compared to the uncorrected scan.
Yes, this is quite good. This is on the edge of what most
people are capable of perceiving.
> BTW: Ed, I do not agree that 3D LUTs are necessarily slow.
There are a lot of pixels when scanning film. Also, it was
the easiest thing for me to incorporate, since VueScan already
used 3x3 matrices everywhere for color correction, especially
for color negative film.
> Therefore I think a profile created with the
> "transparency/positive" setting of a scanner is not representative for
> "transparency/negative" scanning anyway.
The transparency/negative setting on many scanners isn't any
different from the positive setting (in VueScan). On other
scanners, the only difference is that the CCD exposure time
is 2.5 and 3.5 times longer for the green and blue channels. This
can be adjusted for.
Regards,
Ed Hamrick
Bart van der Wolf
"UrbanVoyeur" <nos...@nospam.urbanvoyeur.com> wrote in message
news:3tA5a.855$x....@nwrdny01.gnilink.net...
> Hmm. I not sure I follow you. I am able to embed any of the color
space
> profiles in my tiff files, with or without calibration.
SNIP
After downloading a new version and re-profiling, now Photoshop
recognizes an embedded profile in the scan. Thanks for pointing out
that it did work for you.
Bart
Wolf Faust
wrote:
>"Gerhard Fuernkranz" <nosp...@gmx.de> wrote:
>> Yes, I also fully agree with Wolf, that you won't get too far with a
>matrix
>> only profile, except for scannerns which a priori have a very linear
>response,
>> even without correction.
>
>That's why I modified VueScan 7.6.16 to also do a per-color
>gamma (i.e. power) correction. This worked pretty well.
As said, even if you do optimise the matrix/LUT to the max, you will
not get color faults comparable to typical IT 8 profiling tools. There
are enough books/articles available on this issue.
Note: Lab might not be the best color space for calculating the 3x3
matrix. I think I do have some papers comparing various color spaces
for calculating the 3x3 matrix..
>> With a matrix correction
>> only, I'll never get reasonable results with this scanner. But with a
>simple
>> gamma+matrix profile I can achive an average dE of about 2..3, which isn't
>too
>> bad, compared to the uncorrected scan.
>
>Yes, this is quite good. This is on the edge of what most
>people are capable of perceiving.
Not really. First, this is the mean fault (wich is rather low for a
3x3 matrix), wich usualy means there are a number of color areas that
have a much higher fault. Even small faults in certain areas can cause
huge problems. Take for example the gray patches. A fault even below
dE 1 can be well visible even for the untrained eye. But let's assume
you do get an error of dE 2.3 in one of the gray patches. On your
monitor this fault might not look like a major problem or even kept
unseen. Than you print that image. For the printer this is not a
perfect gray and it might add some color to the gray. Depending on the
printer and printer driver, this can cause extremly distrurbing and
visible faults. I did develop printer drivers for many years and I can
tell you even normal users will find these errors to be unacceptable.
Luckely as they do not see the fault on their monitor, they will most
likely blame the printer manufacturer ;-)
Than you should not forget, that the dE fault is based on a
calculation using the training patches of the target. The real faults
are higher. If you compare current commercial/freeware profilers, they
all have similar mean faults arround dE 1. Yet the quality differences
can be huge as the mean fault of the training patches is a bit
misleading as value for judging the quality. One more realistic
approach is following: a target has 288 patches. Generate 288 profiles
and each time you do remove one target patch from the calibration and
use it as testing patch. Than do not use the mean error. You will
quickly reach a limit where the mean fault will not get better, But if
you look at the standard deviation fault you will see further changes
(hopefully for the better ;-) Of course, generating 288 profiles is
nothing a user can do. But for the programmer of the profiling
software it should be easy to do a loop in his software.
>> BTW: Ed, I do not agree that 3D LUTs are necessarily slow.
>
>There are a lot of pixels when scanning film. Also, it was
>the easiest thing for me to incorporate, since VueScan already
>used 3x3 matrices everywhere for color correction, especially
>for color negative film.
I back up Gerhards comment. A 3D LUT will really only need a couple of
ms more time - not seconds. I think Marti once told me how quick lcms
does the 3D LUT on a decent PC. Really a matter of ms even for large
scans.
But I think you get to the point here. It was easier to implement. ;-)
Ed, the current calibration performance is a big step forward and
nice. We do have lot's of freeware profiling software for scanners and
why reinventing the wheel as you will hardly do any better than for
instance the lcms profiler. So, focus on real important things we
really need and at least change VueScan in such a way that we can use
the other profilers with VueScan. I see two solutions that might be
easy to add:
For instance, check if the scanner profile selected by the user is
from VueScan. If not, apply gamma and convert to 8 Bit before calling
the CMM in order to apply a Lab PCS profile.
The other question is, in a good/typical workflow you usualy do not
alter the scan data until you do output. So how can one embed the
selected custom scanner profile (not just VueScan profiles) to the
scanned file so that the image is loaded correctly by a CMS aware
applications?
Wolf Faust
>what is the current status of the Agfa targets? You wrote on
>your webpage that they would be available at the end of 2002.
As usual, lot's of troubles and various manufacturer orders causing
delays with new stuff. I temporarily stopped the Agfa targets
production also because I had to do some Provia targets for a
manufacturer and I really can only do two films at one time wihtout
causing delays somewhere. And the other more important film was
Velvia. So while the Velvia targets should become available this week,
the Agfa targets will surely take 1-2 months more time before becoming
available.
>Will be target A3 be applicable on CT100 and CT200 Precisa as well?
If I remember correctly yes. But I will look it up again once I
continue working on the Agfa targets...
Ed Hamrick
> As said, even if you do optimise the matrix/LUT to the max, you will
> not get color faults comparable to typical IT 8 profiling tools.
The results in VueScan 7.6.16 are quite good. I scanned the same
Q60 slide on two different scanners and put the results side by
side on the same monitor. I couldn't tell the difference.
> Note: Lab might not be the best color space for calculating the 3x3
> matrix. I think I do have some papers comparing various color spaces
> for calculating the 3x3 matrix..
I'd appreciate it if you could find this and post the best
technique for calculating the 3x3 matrix. VueScan 7.6.16
minimizes the root mean square of dE and weights all patches
the same.
> Not really. First, this is the mean fault (wich is rather low for a
> 3x3 matrix), wich usualy means there are a number of color areas that
> have a much higher fault.
Yes, a few patches have a dE of 10. This is still comparable to
the recommended maximum tolerance of IT8 targets, and as you've
noted elsewhere, it's a better accuracy than the LaserSoft
manufactured IT8 targets ;-)
By far the biggest problem was nonlinearity in the CCD and/or
A/D converter. This is corrected in VueScan 7.6.16.
> I back up Gerhards comment. A 3D LUT will really only need a couple of
> ms more time - not seconds. I think Marti once told me how quick lcms
> does the 3D LUT on a decent PC. Really a matter of ms even for large
> scans.
What is the estimated time per pixel of the cmsDoTransform routine?
> But I think you get to the point here. It was easier to implement. ;-)
Not just ease of implementation - it also helps negative color
correction in VueScan. If you scan the same negative on two different
types of film scanners, you'll get close to the same colors.
> For instance, check if the scanner profile selected by the user is
> from VueScan. If not, apply gamma and convert to 8 Bit before calling
> the CMM in order to apply a Lab PCS profile.
VueScan uses 16 bits per pixel internally.
> The other question is, in a good/typical workflow you usualy do not
> alter the scan data until you do output. So how can one embed the
> selected custom scanner profile (not just VueScan profiles) to the
> scanned file so that the image is loaded correctly by a CMS aware
> applications?
This doesn't work when scanning negatives.
Regards,
Ed Hamrick
Greg
Looking at patch H8, the deltaE is now 15, which is a bit better than it was
before. There is hardly any difference if I select linearize or not. This
still seems far inferior to the LUT profiles I am creating with Little CMS.
I have no idea which patch has the highest dE - I am merely using this one
because it was the one I was using before.
I see that there is an option called "Scanner ICC gamma 2.2". I expected
this to create a raw file with a gamma of 2.2, but the raw file still has a
gamma of 1 if I enable this. Ed - what exactly does this option do? If
Vuescan could output gamma 2.2 (or at least something more perceptually
uniform), I'd be able to use the Little CMS profile checker. (it will be
nice when the LCMS utilities can process high bit depth images).
Greg.
Greg
create an ICC profile with a gamma of 2.2.
So what I've done now is to adjust the gamma of the raw scan to 2.2,
measured it with LCMS' measurement tool, and then used LCMS' profile
checker: max dE is 22.93 (patch F8), but the average is not too bad: 3.97.
Just to make sure, I then went back to Photoshop, and used the raw scan
proper from Vuescan, applying the gamma 1 ICC profile from Vuescan. I looked
at patch F8, and calculated the dE to be essentially exactly the same as
reported by LCMS. (I wasn't sure whether the gamma correction curve I
applied would match the curve which LCMS uses to interpret the gamma in the
ICC file *exactly* - that's why I did this regression test).
Ed: Since I'm still getting large errors, maybe I should give you my raw
scan and also the ICC profile, just to make sure I am not doing something
wrong. Let me know.
Greg.
"Greg" <sulliva...@bigpond.net.au> wrote in message
news:p4k6a.35650$VH5....@news-server.bigpond.net.au...
Erik Krause
> I've fixed both of these things in VueScan 7.6.16, and will
> release it in the next hour or so.
very good! But why did you drop the Image/Slide/Negative curve
options? The log curves gave much better result with negatives...
Regards
Ed Hamrick
> Looking at patch H8, the deltaE is now 15, which is a bit better than it
was
> before.
You'll find that many patches have a dE less than 1.
> There is hardly any difference if I select linearize or not.
It's important for some scanners, not as important for others.
> I see that there is an option called "Scanner ICC gamma 2.2". I expected
> this to create a raw file with a gamma of 2.2, but the raw file still has
a
> gamma of 1 if I enable this. Ed - what exactly does this option do?
It makes the generated ICC file usable with 24-bit raw scan files.
> If
> Vuescan could output gamma 2.2 (or at least something more perceptually
> uniform), I'd be able to use the Little CMS profile checker.
However, these wouldn't be raw scan files any more.
Regards,
Ed Hamrick
Ed Hamrick
> So what I've done now is to adjust the gamma of the raw scan to 2.2,
> measured it with LCMS' measurement tool, and then used LCMS' profile
> checker: max dE is 22.93 (patch F8), but the average is not too bad: 3.97.
Yes, VueScan tries to minimize the root mean square error.
An average dE of 3.97 is good for a matrix profile.
Use the VueScan profile and the LCMS profile and put the two images
side by side. Then ask some neutral person which one looks more like
the paper copy of the IT8 target.
Regards,
Ed Hamrick
Ed Hamrick
> very good! But why did you drop the Image/Slide/Negative curve
> options? The log curves gave much better result with negatives...
VueScan still uses the log (medium) curve for negatives. The
reason I removed these options (and made gamma the default for
image) is that the only option that worked for negatives was
log. You can reproduce the three log settings by changing
"Color|Brightness".
The Color tab was getting unnecessarily crowded.
Regards,
Ed Hamrick
Greg
I'll see if I can do this test. I have what I think is a very good profile
for my Epson 1270.
Please note, however, that if I create a profile using LCMS even for my
cheap flatbed scanner, using the same Provia slide target, the *max* dE of
this profile is less than the *average* dE produced by Vuescan for my Nikon
Coolscan IV film scanner, which is superior to my flatbed scanner. The
average dE of this cheap scanner is only 0.52. Why do you think Vuescan's
results are superior to the LCMS LUT-based profiles?
For my Coolscan IV, LCMS creates a profile with an average dE of 0.37, with
a max dE of 2.23.
Greg.
Greg
> "Greg" <sulliva...@bigpond.net.au> wrote:
> > Looking at patch H8, the deltaE is now 15, which is a bit better than it
> was
> > before.
>
> You'll find that many patches have a dE less than 1.
In fact, only one patch has a dE of less than 1. (patch K11, with a dE of
0.84) I'm still wondering whether I could be doing something wrong, because
you seem to be getting better results.
>
> > I see that there is an option called "Scanner ICC gamma 2.2". I expected
> > this to create a raw file with a gamma of 2.2, but the raw file still
has
> a
> > gamma of 1 if I enable this. Ed - what exactly does this option do?
>
> It makes the generated ICC file usable with 24-bit raw scan files.
>
Ah, so if I select an output resolution of 24-bits for the raw scan, it will
be gamma 2.2? I didn't know that - that's very useful information - thanks.
> > If
> > Vuescan could output gamma 2.2 (or at least something more perceptually
> > uniform), I'd be able to use the Little CMS profile checker.
>
> However, these wouldn't be raw scan files any more.
Agreed. However, I think it would be a useful feature, nevertheless. It
would be easier to use external profiling software in conjunction with
Vuescan, in those cases where the profiling software did not support high
bit depth images. (it would save a gamma adjustment processing step).
Greg.
Greg
my new linearity graph, with correct CIE(Y) data, to
http://users.bigpond.net.au/sullivag/ls40_linearity.png
Greg.
"Greg" <sulliva...@bigpond.net.au> wrote in message
news:nbb6a.26897$VH5....@news-server.bigpond.net.au...
Wolf Faust
wrote:
>"Wolf Faust" <ju...@coloraid.de> wrote:
>> As said, even if you do optimise the matrix/LUT to the max, you will
>> not get color faults comparable to typical IT 8 profiling tools.
>
>The results in VueScan 7.6.16 are quite good. I scanned the same
>Q60 slide on two different scanners and put the results side by
>side on the same monitor. I couldn't tell the difference.
Q60 targets are not working for comparing/testing color faults. Take a
real image prefereable with highly saturated yellow/green and blue.
As said, the error values calculated from a 3x3 matrix with most
scanners is simply too high and will certainly cause visible faults.
Also, a different target or a different scanner can give you very
different (most likely higher). faults. I expect scanners with a light
source that are far from monochromatic will have bigger problem than
those with a more monochromatic light source. If you do have a
monochromatic light source matching the film dyes in wavelength than a
3x3 matrix should be all you need to correct the scan. But most of us
have to live with the consumer scanners one can afford :-(
Don't forget other problems like the one I described regarding the
printing faults.
>Yes, a few patches have a dE of 10. This is still comparable to
>the recommended maximum tolerance of IT8 targets, and as you've
dE 10 is the max deviation an IT 8.7 target patch may have compared
to the values in the standard. 99% of the patches of one target must
have an error below dE 5 compared to the reference file - wich is
basicly the reproduction fault. Good IT 8 targets do have a mean
reproduction fault of dE 0.2-0.6. Manufacturers are required to report
the standard deviation of their reproduction faults and you can find
this and other info in my target reference files.
So good targets are really beyond any doubts and should not cause any
visible faults. The faults of the matrix profile are MUCH higher in
comparison.
I think the 3x3 matrix is a big step. The missing feature is, that you
can't use other scanner profiles than VueScans.
>> For instance, check if the scanner profile selected by the user is
>> from VueScan. If not, apply gamma and convert to 8 Bit before calling
>> the CMM in order to apply a Lab PCS profile.
>
>VueScan uses 16 bits per pixel internally.
Yes. wich is why I thought you should convert before calling the CMM
if it is not a VueScan profile. The main issue is getting VueScan to
work with other scanner profiles. After the profile was applied you
can convert back to 16 Bit if necessary so that VueScan can continue
working with the data.
>This doesn't work when scanning negatives.
This is not intended (yet ;-) Correct me if I am wrong: currently you
have to convert the scan to an output device during scanning or use
ananother tool to embed the scanner profile to the raw scan in order
to get a scan with embedded profile?
Ed Hamrick
> If you do have a
> monochromatic light source matching the film dyes in wavelength than a
> 3x3 matrix should be all you need to correct the scan. But most of us
> have to live with the consumer scanners one can afford :-(
That's one reason I like scanners with LED light sources - the
colors are often quite good with 3x3 matrix correction.
The Nikon film scanners and Canon LIDE scanners both use LED
light sources.
> I think the 3x3 matrix is a big step. The missing feature is, that you
> can't use other scanner profiles than VueScans.
I improve VueScan one feature at a time.
> Yes. wich is why I thought you should convert before calling the CMM
> if it is not a VueScan profile. The main issue is getting VueScan to
> work with other scanner profiles. After the profile was applied you
> can convert back to 16 Bit if necessary so that VueScan can continue
> working with the data.
This would lose bits of precision.
> >This doesn't work when scanning negatives.
>
> This is not intended (yet ;-) Correct me if I am wrong: currently you
> have to convert the scan to an output device during scanning or use
> ananother tool to embed the scanner profile to the raw scan in order
> to get a scan with embedded profile?
VueScan works fine scanning color negative film using a 3x3 matrix
profile generated from an Ektachrome targets. It isn't trivial to
make this work, but it does work.
I use the 3x3 matrix to convert to a color space that matches
(to the degree possible) the colors of the negative film dyes.
It's not perfect, but the images look good.
Regards,
Ed Hamrick
Greg
> The results in VueScan 7.6.16 are quite good. I scanned the same
> Q60 slide on two different scanners and put the results side by
> side on the same monitor. I couldn't tell the difference.
When I look at the colour differences in the Little CMS profile checker,
there are *very* easily visible faults. And this is on a standard monitor,
which has quite a small gamut. If the results from different scanners are
consistent, that's good, but it doesn't mean that the results are accurate.
:)
I don't mind a *bit* of colour error. (not as much as what is currently
being produced by Vuescan's profiles, though) But I do mind if I'm not
capturing the gamut of the image, and I do not think that Vuescan is doing
that at the moment, if my interpretation of the following chromacity error
plot is correct:
http://users.bigpond.net.au/sullivag/xy.png This shows the colour samples of
the IT8 test target, along with associated colours produced by Vuescan's
profile. I want the extra gamut so I can get the most out of my Epson 1270
printer, and also for archiving, so I can make use of the extra gamut later,
as display (and printer) technology improves. It is a *lot* of work to scan
film, so I want to do it properly.
For the time being, I will use an external profiling solution when I use
Vuescan, because I do not think that Vuescan's current profiling is good
enough. Since Vuescan is the best program I have available to scan
negatives, and I can't profile negatives, I have no choice but to use
Vuescan, but from now on I will be saving the raw scans, so that any future
improvements in Vuescan will allow me to go back and re-process. To date, I
have not been archiving the raw scans, and I do not have confidence that
Vuescan has captured the full gamut properly in the processed scans (with
embedded "file" colour space.). But I agree totally that these scans look
good on my monitor.
Greg.
Erik Krause
you wrote...
> VueScan still uses the log (medium) curve for negatives. The
> reason I removed these options (and made gamma the default for
> image) is that the only option that worked for negatives was
> log. You can reproduce the three log settings by changing
> "Color|Brightness".
But I can't use log for slide any more if I want...
> The Color tab was getting unnecessarily crowded.
That's why I suggested to put the IT8 stuff into a wizard. The only
option really needed is 'built-in' or 'ICC' and file name. Perhaps it
would be a good idea to place film base color values and may be film
types on the device tab since I have to choose/lock it there anyway.
Or move some points to the filter tab (restore color/fading are color
options, too)...
--
Erik Krause
Digital contrast problems: http://www.erik-krause.de/contrast