Profile Prism frustration

[J A K]

I'm looking forward to testing out the GretagMacbeth
Spectrophotometer and their super duper profiling software. While
I'm not sure about specialty papers and/or inks; I have a real hard
time believing the GretagMacbeth setup will outperform the profile
I have for my plain Jane Epson 1270 and Epson's HW Matte paper
using Profile Prism.

I'll bet they will!

I’m going to have to put you at the end of the betting line (my window is full). I think my buddy (the guy that owns the GretagMacbeth hardware and software) also wants a piece of the action. For now I’d like to check out his hardware and get an understanding of the profiling software before any hard cash is up for grabs. Money aside, for my specific printer and the HW Matte paper, I’ll be extremely impressed if the GretagMacbeth system can create a profile as good as the one I just made with PP.

Send me a target (Avaiable on my site) and I'll
make you a profile if you agree to report back here what you find.
You'll need to output a standard test image (there is one on my web
page). I only use the new TC 9.18 target for ink jet profiles.

Andrew – thanks for your generous offer. Yes – I would REALLY like to take you up on the evaluation profile offer, and I will definitely report the results to the forum. However, I would appreciate a short-term rain check. My immediate problem is schedule. I’m still using the old Epson inks (the made in Japan version that is no longer available). I’m just about out (circa Dec 2002 or Jan 2003) and will be purchasing a large quantity of the new inks (now made in China). If there is going to be a profile created by a real pro (the Digital Dog), I would like it to be with the new Epson inks. Is the rain check okay with you? If not, I’ll just toss out my last 1.5 cartridges (they expired a year ago) and get some of the new inks and print out your targets (after super flushing the print heads). Please let me know what your preference is.

BTW, I already have i1 TC 9.18 part 1 and i1 TC 9.18 part 2. However, just to double check, I went to your site and downloaded your ZIP file. Turns out that your 9.18 targets look completely different than mine and have a much larger file size (758 Kb vs. 3478 Kb). Regarding the differences, I'm more than a little confused. Have you customized the GretagMacbeth targets for your profile creation process?

A few things to look at (my test image has the areas): Spectral
gradient. Is it smooth and do the blues look pure or shiift to
cyan/magenta (common with profiles)?

The "blues" look smooth and pure - there is no spectral shift in the blue wedge. It is interesting that out of all the wedges (gray, red, green, blue, magenta, cyan, and yellow), if there is even a slight hint of a color shift it is in the gray wedge. Obviously, I really should measure the chromaticity coordinates of each step rather than go by my eyeballs. Hopefully, I can get my test print measured soon.

Are all the steps neutral and
as you've seen, separation on 21 steps.

Sorry - I should have been a little more specific about the "21 shades" statement. In my "rush to profile happiness", I was discussing the number of steps only in the gray wedge and did not make that clear. Let me do the job right this time and list all colors wedges individually.

Gray and yellow: 21 steps
Cyan, magenta: 20 steps
Red: 19 steps
Blue: 10 or 11 steps (really too close to call)

My numbers are not a surprise – compared to other colors, a larger x-y coordinate shift in the blue region is necessary to discern a change in color.

The spectrophotometer seems to have a few issues with reading some
watercolor paper - GretagMacbeth technical support folks have not
come up with a solution.

Which spectro?

The system is based on an Eye-1 Pro, Profile Maker 4.1, Wasatch Soft RIP 4.5, and an HP5000 printer with UV inks. My buddy uses Photoshop 6.01 and I use Photoshop 7.01 (I can’t imagine the Photoshop version is related to his problem). Anyway, the primary problem with the watercolor paper(s) is maximizing gamut size without compromising spatial resolution. My friend knows who you are; he was thinking about going up to New Mexico for one of your classes (and dragging me along for company). It is awful cold in New Mexico this time of the year – are you going to offer classes in Phoenix, AZ any time soon?

Is there anything special you do when it
comes to reading watercolor paper with your spectrophotometer (if
memory serves me correctly, you have a GretagMacbeth
spectrophotometer)?

I have several. I've just started using the new iCColor instrument
which does multiple samples per patch to average. That's necessary
with papers that have a surface that isn't smooth. I was using a
Spectroscan (still have and love it) and it takes 45 minutes to
read the TC9.18 target with ONE patch reading. iCColor takes 10 to
do the same job with multiple reads per patch. I also use the
EyeOne on location. It reads 100 samples per second!

Cool deal – the EyeOne sounds like the right horse for the races. How does the EyeOne accuracy compare to the other spectrophotometers you own?

Again, thanks for the offer and especially for taking time to answer my questions.

Sincerely,

Joe Kurkjian, Pbase Supporter

http://www.pbase.com/jkurkjia

 

[Mike Chaney]

And the printer you wish to profile. Some printers simply can't be
profiled using a scanner (nature of the inks and papers play a
role).

A scanner is NOT a spectrophotometer which is really needed to
build an ICC profile. Using a scanner can work great in some cases
and awful in others. There have been several scanner based products
on the market for years (Wiziwyz, MonacoEZ etc) and their sucess
has been hit or miss. With pigmented inks (especially something
like an Epson 2000p), scanners fail big time in provding the data
necessary to make a lot of educated guesses to take RGB data and
produce a profile where spectral data is really necessary.

This is true to a certain extent, but many of the better scanners if set up properly, can perform nearly as well as a spectrophotometer. The scanner is actually more of a sore spot than the printer as there are some scanner models that just are not up to the task of being used as a makeshift spectro. The problem here, and the reason that so many professionals will tell you that scanner based printer profiling is hit and miss has to do with the following factors:

(1) The software you are using to profile. Profile Prism is the only scanner based printer profiling package on the market that gives you the information and tools you need to be able to utilize your scanner as a makeshift spectrophotometer. The other tools just have you place targets on the flat bed and click the scan button. That will never work on any scanner! In their default scanning mode, all scanners will clip a good portion of the shadows and highlights because they try to auto-contrast and expose for the "pop" tha most people like to see in images. If set up properly (and the Profile Prism documentation shows you how to do this), the scanner will actually perform quite well for this task, but you have to know how to get "raw" scans to do this.

(2) The scanner itself. Some scanners just don't perform well as spectrophotometers and don't come with software/hardware that allow you to get a truly successful scan. I've seen models that will clip the scans no matter what you do, and others (like most of the Epsons) that do a great job. In this case, the scanner can actually prevent you from getting a good profile. Again, unlike other tools which happily build useless profiles and never give you any information about the quality of the scan you are using, Profile Prism can almost always detect when a scan is "too bad" to be used for profiling. This helps cut down on the instances where users build profiles "blindly" and end up with horrible results.

(3) The paper and metamerism. Many people try to use papers that are not designed for their printer and while not obvious by looking at the printouts, the printer is actually not working properly on that paper. This can cause problems with many profilers because ink is not distributed properly, causing very abrupt color shifts that cannot be corrected. In addition, many papers are not white and are actually blue/cyan due to brighteners. This, combined with the fact that some (particularly dye) inks have metamerism issues under different light sources, can cause a problem. Unfortunately, this is not a problem that can be completely "fixed" just by inserting a spectrophotometer into the equation either!

(4) Gamut mapping. The current color management model (CMM) used for ICC 2.0 is actually quite deficient. Because the CMM has no knowledge of the gamut of the input image versus the gamut of the output profile, many compromises must be made to make your image appear to look reasonable. In addition, the gamut of most printers in some areas (like blue for example) is so shallow that again, compromises must be made to make the print "appear" correct to the eye when in fact, there is no way that the printer can print the specified color. Because it is physically impossible to reproduce some colors that may be in your images, the appearance of "correct" color will vary by the amount of compromise being made by the gamut mapping. This means that some papers will look better than others. If you use a paper with small dynamic range and which has a smaller gamut for the given inks, it may not look as good as a higher quality paper.

FWIW, I've heard from a number of people who have successfully profiled numerous papers on dye based printers like the Epson 200P using Profile Prism and their desktop scanner. That's why I hesitate to agree with the statement that some printers just cannot be profiled with a scanner. I'd rather say that there are some printer/scanner combinations that don't work well and will not be successful in being used in tandem to create profiles. I've also heard from a few who were not successful with the 2000P. Again, I think it depends a lot on the scanner being used. The harder the printer is to profile, the better scanner you need to get the best results.

There's no doubt that a spectrophotometer will do a better job than a scanner at profiling, but if you take the time to address all the variables in using your scanner to do the job (depending on the scanner), the difference is not very noticeable as I've done direct comparisons myself.

--
Mike
http://www.ddisoftware.com

 

[ajs2107922]

Slightly of topic but mike what scanner would you recommend. I have an Epsom 1280(1290) printer whats best?

And the printer you wish to profile. Some printers simply can't be
profiled using a scanner (nature of the inks and papers play a
role).

A scanner is NOT a spectrophotometer which is really needed to
build an ICC profile. Using a scanner can work great in some cases
and awful in others. There have been several scanner based products
on the market for years (Wiziwyz, MonacoEZ etc) and their sucess
has been hit or miss. With pigmented inks (especially something
like an Epson 2000p), scanners fail big time in provding the data
necessary to make a lot of educated guesses to take RGB data and
produce a profile where spectral data is really necessary.

This is true to a certain extent, but many of the better scanners
if set up properly, can perform nearly as well as a
spectrophotometer. The scanner is actually more of a sore spot
than the printer as there are some scanner models that just are not
up to the task of being used as a makeshift spectro. The problem
here, and the reason that so many professionals will tell you that
scanner based printer profiling is hit and miss has to do with the
following factors:

(1) The software you are using to profile. Profile Prism is the
only scanner based printer profiling package on the market that
gives you the information and tools you need to be able to utilize
your scanner as a makeshift spectrophotometer. The other tools
just have you place targets on the flat bed and click the scan
button. That will never work on any scanner! In their default
scanning mode, all scanners will clip a good portion of the
shadows and highlights because they try to auto-contrast and expose
for the "pop" tha most people like to see in images. If set up
properly (and the Profile Prism documentation shows you how to do
this), the scanner will actually perform quite well for this task,
but you have to know how to get "raw" scans to do this.

(2) The scanner itself. Some scanners just don't perform well as
spectrophotometers and don't come with software/hardware that allow
you to get a truly successful scan. I've seen models that will
clip the scans no matter what you do, and others (like most of the
Epsons) that do a great job. In this case, the scanner can
actually prevent you from getting a good profile. Again, unlike
other tools which happily build useless profiles and never give you
any information about the quality of the scan you are using,
Profile Prism can almost always detect when a scan is "too bad" to
be used for profiling. This helps cut down on the instances where
users build profiles "blindly" and end up with horrible results.

(3) The paper and metamerism. Many people try to use papers that
are not designed for their printer and while not obvious by looking
at the printouts, the printer is actually not working properly on
that paper. This can cause problems with many profilers because
ink is not distributed properly, causing very abrupt color shifts
that cannot be corrected. In addition, many papers are not white
and are actually blue/cyan due to brighteners. This, combined with
the fact that some (particularly dye) inks have metamerism issues
under different light sources, can cause a problem. Unfortunately,
this is not a problem that can be completely "fixed" just by
inserting a spectrophotometer into the equation either!

(4) Gamut mapping. The current color management model (CMM) used
for ICC 2.0 is actually quite deficient. Because the CMM has no
knowledge of the gamut of the input image versus the gamut of the
output profile, many compromises must be made to make your image
appear to look reasonable. In addition, the gamut of most printers
in some areas (like blue for example) is so shallow that again,
compromises must be made to make the print "appear" correct to the
eye when in fact, there is no way that the printer can print the
specified color. Because it is physically impossible to reproduce
some colors that may be in your images, the appearance of "correct"
color will vary by the amount of compromise being made by the gamut
mapping. This means that some papers will look better than others.
If you use a paper with small dynamic range and which has a smaller
gamut for the given inks, it may not look as good as a higher
quality paper.

FWIW, I've heard from a number of people who have successfully
profiled numerous papers on dye based printers like the Epson 200P
using Profile Prism and their desktop scanner. That's why I
hesitate to agree with the statement that some printers just
cannot be profiled with a scanner. I'd rather say that there are
some printer/scanner combinations that don't work well and will not
be successful in being used in tandem to create profiles. I've
also heard from a few who were not successful with the 2000P.
Again, I think it depends a lot on the scanner being used. The
harder the printer is to profile, the better scanner you need to
get the best results.

There's no doubt that a spectrophotometer will do a better job than
a scanner at profiling, but if you take the time to address all the
variables in using your scanner to do the job (depending on the
scanner), the difference is not very noticeable as I've done direct
comparisons myself.

--
Mike
http://www.ddisoftware.com

--
Alex
LWS photographic (UK)

 

[Andrew Rodney]

Is the rain check okay with you?

Sure but put a note in when you send the target (perhaps print out one of these threads) so I know what the story is.

BTW, I already have i1 TC 9.18 part 1 and i1 TC 9.18 part 2.
However, just to double check, I went to your site and downloaded
your ZIP file. Turns out that your 9.18 targets look completely
different than mine and have a much larger file size (758 Kb vs.
3478 Kb).

There are several flavors of this target depending on the device used to measure them. The two page file on my site if for the new GretagMacbeth iCColor spectro. There is a 2nd file there (optional for devices that can only output an 8x10 bleed) that is the same target but for the Spectroscan. You can use either but the two page is preferred because it's way faster and the iCColor does a lot of patch averaging when doing the measure.

Regarding the differences, I'm more than a little
confused. Have you customized the GretagMacbeth targets for your
profile creation process?

No although that is possible with this very robust package. I did size the TC.9.18 two page target down a bit so it would fit on an 8.5x11 page which is so comon.

A few things to look at (my test image has the areas): Spectral
gradient. Is it smooth and do the blues look pure or shiift to
cyan/magenta (common with profiles)?

Just so we are on the same page, I view all prints under a GTI D50 light box with dimmer. Also, ProfileMaker Pro allows me to build profiles for multiple lighting conditions! Very cool feature. I build the profiles using the GTI lightbox setting in the software (rather than the default D50 viewing light). The differences are not huge but I prefer the GTI setting. Bottom line is you just have to have a D50 light box to view the prints under.

Which spectro?

The system is based on an Eye-1 Pro, Profile Maker 4.1,

The EyeOne is a great spectro. I use it on location because it's small and fast. I seriously doubt you'll see any difference in the EyeOne verses the iCColor.

My friend knows who you
are; he was thinking about going up to New Mexico for one of your
classes (and dragging me along for company). It is awful cold in
New Mexico this time of the year – are you going to offer classes
in Phoenix, AZ any time soon?

My class at the Santa Fe workshops is in April. Weather should be lovely.

Is there anything special you do when it
comes to reading watercolor paper with your spectrophotometer (if
memory serves me correctly, you have a GretagMacbeth
spectrophotometer)?

Having more than one reading pre patch is important for sure due to the surface. Not an issue with EyeOne (100 measurements per second).

--
Andrew Rodney
http://www.digitaldog.net

 

[Mike Chaney]

Slightly of topic but mike what scanner would you recommend. I have
an Epsom 1280(1290) printer whats best?

Just about any Epson model starting with the 1250/1260 and up gives you the most bang for the buck in being able to have a scanner that can deliver accurate profiles. This is partly due to the hardware and partly due to the excellent Twain driver that the Epsons come with. The 2400/2450 is one of the best I've seen but the difference between those models and the lower ones doesn't usually justify the price difference for most people.

HP scanners also do a fine job, but there's a bigger quality spread on their models. I've had the least luck profiling with Canon scanners. They all seem to have trouble with clipping and all want to overcontrast the scans. For some reason, this is difficult to overcome with the scanning software that comes with the Canons.

Just my .02.

--
Mike
http://www.ddisoftware.com

 

[Andrew Rodney]

This is true to a certain extent, but many of the better scanners
if set up properly, can perform nearly as well as a
spectrophotometer.

As well? I find that hard to believe but I see no reason to argue. If you had a Mac version, I'd love to do a side by side test to a few printers. I should say I HAVE seen some output devices profiled with a scanner based package produce results that were very, very close to using a spectro. My Fuji PG3000 is such a device. The older Epson 2000P was a disaster.

Spectral data is important for a lot of reasons such as the ability to detect and account for optical brighteners which ProfileMaker Pro does when feed spectral data.

The scanner is actually more of a sore spot
than the printer as there are some scanner models that just are not
up to the task of being used as a makeshift spectro.

I agree. When I tested the old Praxisoft Wiziwyg using both a $18K Imacon in reflective mode and a $150 Agfa Snapscan, the differences in profile quality was moot. But neither could build a decent profile for the 2000P or other devices that have tricky inks/media/substraights. This was true with MonacoEZ as well.

The problem here, and the reason that so many professionals will tell you that
scanner based printer profiling is hit and miss has to do with the
following factors:

(1) The software you are using to profile. Profile Prism is the
only scanner based printer profiling package on the market that
gives you the information and tools you need to be able to utilize
your scanner as a makeshift spectrophotometer.

With all due respect, until a scanner can read spectral data, it's not a Spectrophotometer. You (your software) has to make a lot of guesses about the data being feed to end up with some spectral recipe to build a profile. You may be doing this MUCH better than anyone else but let's be realistic here about what a scanner can and cannot do compared to a Spectrophotometer.

The other tools
just have you place targets on the flat bed and click the scan
button. That will never work on any scanner! In their default
scanning mode, all scanners will clip a good portion of the
shadows and highlights because they try to auto-contrast and expose
for the "pop" tha most people like to see in images.

By and large this somewhat true but not something I think you should put in stone. The way you setup the scanner is very important and having feedback to the right settings is an excellent idea. But again, with other products, I've had very good results (again, to my PG3000) and results that were just awful using the same scanners and software. No amount of scanner tweaking is going fix the issues.

(2) The scanner itself. Some scanners just don't perform well as
spectrophotometers and don't come with software/hardware that allow
you to get a truly successful scan.

No scanner is a spectrophotometer. Let's be clear. You're getting RGB data, not spectral data. You have to make assumptions about the data you’re getting.

(3) The paper and metamerism. Many people try to use papers that
are not designed for their printer and while not obvious by looking
at the printouts, the printer is actually not working properly on
that paper. This can cause problems with many profilers because
ink is not distributed properly, causing very abrupt color shifts
that cannot be corrected.

I've yet to run into a paper that I couldn't profile using the right hardware and software but I would agree that for a scanner based profile package, this is true.

In addition, many papers are not white
and are actually blue/cyan due to brighteners.

And my Spectro can "see" this and my software can account for it.

Unfortunately,
this is not a problem that can be completely "fixed" just by
inserting a spectrophotometer into the equation either!

Completely? No. But I can build an output profile for various light sources using spectral data that knows about both UV issues in the paper and how the colors should (I did say should) react to the light source. It's not perfect but I know the next version of the software will have even more tools to account for theses issues. But the software HAS to have spectral data (which is now the default when measuring) to account for this.

(4) Gamut mapping. The current color management model (CMM) used
for ICC 2.0 is actually quite deficient. Because the CMM has no
knowledge of the gamut of the input image versus the gamut of the
output profile, many compromises must be made to make your image
appear to look reasonable.

VERY true. And there is no spec what so ever for how a software manufacturer can deal with a perceptual intent (hence the reason I find a RelCol intent usually produces better results). At a recent CMS Shoot out at Seybold I did with Bruce Fraser, Steve Upton and Chris Murphy, we found some pretty funky assumptions being made when using Percetual rendering intents (and in some cases other intents). Gamup mapping is certainly an art!

FWIW, I've heard from a number of people who have successfully
profiled numerous papers on dye based printers like the Epson 200P
using Profile Prism and their desktop scanner.

Well I'd love to do a test... And I think you'd do well to go after the "big boys" and come out with an OSX version of your software. I think a product in your price range is necessary for the market.

I've also heard from a few who were not successful with the 2000P.
Again, I think it depends a lot on the scanner being used.

Do you list scanners that you know (recommend) will work?

There's no doubt that a spectrophotometer will do a better job than
I've done direct comparisons myself.

I'm sure you have but I haven't .
--
Andrew Rodney
http://www.digitaldog.net

 

[Mike Chaney]

Spectral data is important for a lot of reasons such as the ability
to detect and account for optical brighteners which ProfileMaker
Pro does when feed spectral data.

I can see where spectral data could allow you to address metamerism and better handle the paper white point, BUT... profiles cannot address spectral data in any way, shape or form. In the end, your profile is going to be operating in either Lab or XYZ color spaces, neither of which contains any spectral information.

What this means is that while a spectral analysis can help you overcome some metamerism and paper white point issues by allowing you to choose the light source that the profile works for, it will still only be valid for a single light source. In contrast, the scanner profile is really only accurate for the light produced by the scanner lamp.

So I agree spectral data gives you more flexibility, but it does not necessarily make your profile more "accurate". It just means you can better choose your light source that works best with the profile. Spectral data does you no good other than that purpose because ultimately your ICC profiles have to be built on the least common denominator (the CMM)... so you basically have to throw away that spectral information anyway before building your profile because your profiles cannot account for it due to the limitations of the PCS.

--
Mike
http://www.ddisoftware.com

 

[Andrew Rodney]

I can see where spectral data could allow you to address metamerism
and better handle the paper white point, BUT... profiles cannot
address spectral data in any way, shape or form. In the end, your
profile is going to be operating in either Lab or XYZ color spaces,
neither of which contains any spectral information.

Ah but what data is needed to BUILD the profile?

What this means is that while a spectral analysis can help you
overcome some metamerism and paper white point issues by allowing
you to choose the light source that the profile works for, it will
still only be valid for a single light source. In contrast, the
scanner profile is really only accurate for the light produced by
the scanner lamp.

That could be a real problem no?

So I agree spectral data gives you more flexibility, but it does
not necessarily make your profile more "accurate".

No but don't you eventually need either spectral data or some device independat colorspace (LAB) receipe to build the profile? One device can provide nothing more than RGB data and the other can produce spectral data. At some point, don't you need the latter just to build the profile? If so, isn't some assumptions being made about the conversion from RGB to LAB?

--
Andrew Rodney
http://www.digitaldog.net

 

[J A K]

Is the rain check okay with you?

Sure but put a note in when you send the target (perhaps print out
one of these threads) so I know what the story is.

Thanks!

Also, ProfileMaker Pro allows me to build
profiles for multiple lighting conditions! Very cool feature.

Wow - that really is cool.

My friend knows who you
are; he was thinking about going up to New Mexico for one of your
classes (and dragging me along for company). It is awful cold in
New Mexico this time of the year – are you going to offer classes
in Phoenix, AZ any time soon?

My class at the Santa Fe workshops is in April. Weather should be
lovely.

Thanks - I'll pass that information along. In case I do tag along with my buddy, I need to ensure the information is kept from my wife. We lived in Albuquerque for eight years and she just loves to go shopping in Santa Fe.

Well - I think we are all set. Thanks again for all your help/comments and the profile offer; the help you have provided is very much appreciated.

Regards,

Joe Kurkjian, Pbase Supporter

http://www.pbase.com/jkurkjia

 

[Mike Chaney]

address spectral data in any way, shape or form. In the end, your
profile is going to be operating in either Lab or XYZ color spaces,
neither of which contains any spectral information.

Ah but what data is needed to BUILD the profile?

If you know the intended viewing light source, you could use the spectral data to build a better profile that is better under that light source. That would give you the proper Lab --> RGB matrix for that light source, but when you actually build the profile, you've already omitted the spectral data by that point and converted to Lab.

still only be valid for a single light source. In contrast, the
scanner profile is really only accurate for the light produced by
the scanner lamp.

That could be a real problem no?

Yes. It is particularly problematic when dealing with inks that have significant metamerism and when the scanner lamp is different from the intended viewing conditions. Here, a spectro really helps. Even still, most scanner lamps are in the 5000K range and while not "full spectrum", they usually do a good job for some of the most common viewing conditions. Ability to design a profile for a specific light source is much more useful when the inks you are using suffer from metamerism though. Sometimes you know the light source that the print will be subject to, and in those cases, I can see where a spectro would be much better than a scanner. If you are trying to produce a print that you'd like to look reasonable under an arbitrary light source such as you might when sending photos to clients, I don't really see where it matters because why choose a specific light source when the assumption you make is probably going to be no better than whatever your scanner threw at it. They are both arbitrary in that case.

So I agree spectral data gives you more flexibility, but it does
not necessarily make your profile more "accurate".

No but don't you eventually need either spectral data or some
device independat colorspace (LAB) receipe to build the profile?
One device can provide nothing more than RGB data and the other can
produce spectral data. At some point, don't you need the latter
just to build the profile? If so, isn't some assumptions being made
about the conversion from RGB to LAB?

Not really. RGB data is really no different with regard to "implementation" than a device independent color space like Lab or XYZ which is used internally by profiles. The only difference is that Lab and XYZ are known/static spaces and they are wider than a device dependent color space. Still, device dependent RGB, Lab, and XYZ all operate on the same principles and the same math. They are simply different color spaces, and none of them can address spectral response.

To look at it another way, your profile is nothing more than a look up table (LUT) that maps Lab values to the device dependent RGB values needed for your printer. Before you can even start to build a profile, you've already had to take spectral data and convert it to Lab color space based on an assumption for light source, and in doing so, you've already eliminated your spectral data at step 1.

The way I see it, spectral data can be used up front to build a profile that is [potentially much] better if you know your viewing conditions. If you don't, and/or the ink you use doesn't suffer from noticeable metamerism, I think the point is moot. It's like buying tires for a car. You could buy an expensive machine that reads the road surface and tells you exactly what kind of tire to put on the car for the best performance on that surface. The problem is, the car can only have one set installed at any one time, and if you don't know what road surfaces you'll encounter in your trip, the usefulness of that machine is debatable.

--
Mike
http://www.ddisoftware.com

 

[Andrew Rodney]

If you know the intended viewing light source, you could use the
spectral data to build a better profile that is better under that
light source. That would give you the proper Lab --> RGB matrix
for that light source, but when you actually build the profile,
you've already omitted the spectral data by that point and
converted to Lab.

But don't you have to start with device dependant RGB from the scanner to end up with LAB to produce the profile?

That could be a real problem no?

Yes. It is particularly problematic when dealing with inks that
have significant metamerism and when the scanner lamp is different
from the intended viewing conditions.

The question then begs the answer, how often does that happen?

still, most scanner lamps are in the 5000K range and
while not "full spectrum", they usually do a good job for some of
the most common viewing conditions.

So are you assuming 5000K or building a scanner profile first (ala MonacoEZ)?

you are trying to produce a print that you'd like to look
reasonable under an arbitrary light source such as you might when
sending photos to clients, I don't really see where it matters
because why choose a specific light source when the assumption you
make is probably going to be no better than whatever your scanner
threw at it. They are both arbitrary in that case.

While ProfileMaker Pro is somewhat unique in allowing users to specify a viewing condition, I think most profile packages assume the image will be viewed under a D50 light box (D65 overseas?). So I'm cool with that basic assumption for now.

So I agree spectral data gives you more flexibility, but it does
not necessarily make your profile more "accurate".

Here is where I'm trying to understand how you take device dependant RGB from an unprofiled (?) scanner and end up with the necessary information to build an ICC Profile. Do you not need either LAB or spectral data to make this happen? If so, how do you go from RGB to this space (and I'm not asking for any special trade secrets here, just whether or not you have to make a lot more assumptions from the RGB you get).

To look at it another way, your profile is nothing more than a look
up table (LUT) that maps Lab values to the device dependent RGB
values needed for your printer.

That I understand. I'm trying to figure out how you take the RGB and use it in a fashion that mimics what a Spectro that cost a heck of a lot more does. Why is it that so few packages (2 that I know of today) use a scanner?

Before you can even start to build
a profile, you've already had to take spectral data and convert it
to Lab color space based on an assumption for light source, and in
doing so, you've already eliminated your spectral data at step 1.

OK but you had spectral data or some LAB vaules at some point. How did the come about from RGB data off a scanner?

No comment about being platform agnostic?
--
Andrew Rodney
http://www.digitaldog.net

 

[pnthobby]

I thought I recognized that post.... Sorry buddy, can't help you on that account, but I'm just trying to get a decent photo printer for output. Nothing as nice as your stuff! By the way, I see next month on the calendar (VRCC) that your beauty is there! Way to go! See you soon???

Paul Weber
--
'No matter where you go, there you are...' Buckaroo Banzai

 

[Mike Chaney]

But don't you have to start with device dependant RGB from the
scanner to end up with LAB to produce the profile?

Yes. The key here is that you scan two targets at the same time. You scan a known/calibrated target with your printed target in the same scan. If you do that, you can take those results (the scanner profile) and easily develop a profile that takes the scanner's RGB values and converts the scanned data directly to Lab color space. After you do that, you examine the printed target that was scanned with your reference and you can derive the Lab values for each printed color patch.

Where a spectro steps ahead of a scanner is that the spectro captures the spectral info directly. You can then take that (more accurate and more direct) spectral info and calculate your Lab color values directly from the spectral info. It's a bit more direct than using a calibrated scanner and should be more accurate. It also allows you to select your light source which you can't do if you start with a composite (RGB).

The interesting thing about the way scanners and spectros build profiles is that they both eventually end up at the same endpoint At some point in the process, you need to convert spectral info into "dominant wave length" that can be expressed as a composite (either RGB or Lab) that contains no spectral info. The difference is the point at which this conversion is done. The scanner drops the spectral info and creates a composite at the time of the scan. I'm sure this is less accurate than a spectro which records the actual spectral info and then converts that to a composite, dropping the spectral info as part of that mathematical conversion.

Keep in mind that I'm not trying to sell scanner profiling on people who have spectros. A spectro will always be more accurate than a scanner and will (should) always do a better job at profiling. The issue that I'm addressing is that there is something inherently wrong with scanners that make them insufficient for the task of being used as a [makeshift] spectrophotometer. It does depend on the conditions, but in many cases, they perform quite well and certainly sufficient enough to be able to profile most desktop scanners and create profiles with reasonable accuracy.

Yes. It is particularly problematic when dealing with inks that
have significant metamerism and when the scanner lamp is different
from the intended viewing conditions.

The question then begs the answer, how often does that happen?

In my experience, not very often. The exception is that I have found a number of old (> 3 years old) scanners that have lamps that produce light that is very different than most modern models. Also, inks with metamerism are more difficult to profile, but still possible under many conditions.

So are you assuming 5000K or building a scanner profile first (ala
MonacoEZ)?

Yes. That's the only way to reliably build a printer profile. The raw scan must be used to profile the scanner using a known/calibrated target, and that scanner profile is then used to derive RGB --> Lab (or Lab --> RGB depending on which side of the fence you are on). The two (known/calibrated target and printed target) must be scanned together in the same scan. Scanning both together allows you to develop a relationship between the known target and the printer target and that gives you the ability to derive (D50 based) Lab values for all the printed color patches. In turn, that gives you your printer profile.

Here is where I'm trying to understand how you take device
dependant RGB from an unprofiled (?) scanner and end up with the
necessary information to build an ICC Profile. Do you not need
either LAB or spectral data to make this happen?

Same answer as above. You can't do it (properly) with a raw scan and no calibration target. If you start with a scanner that is set up properly to produce a non-clipped, raw scan, the fact that you scanned a known target along with your printer target gives the profiler the ability to: (1) profile the scanner first and then (2) use that scanner profile to derive Lab values for the color targets printed on the printer target.

That I understand. I'm trying to figure out how you take the RGB
and use it in a fashion that mimics what a Spectro that cost a heck
of a lot more does. Why is it that so few packages (2 that I know
of today) use a scanner?

I think the answer to that is simple. To really do it right (with a scanner), you have to be meticulous in how you make your scans and follow some very specific instructions. You also need some feedback from the software and the software has to be smart enough to tell you when the scan is bad I guess most companies feel that it is beyond most users or they don't want to deal with the issue of helping people get scans that are good enough to use for profiling. Read the page below and you'll start to see why: (1) scanner based profiles don't work for a lot of people using the packages with no feedback and (2) why some companies simply might not want to deal with it:

http://www.ddisoftware.com/prism/help/psteps.htm

I'm comfortable saying that it is more difficult to get good results with a scanner based profiler when compared to using a spectro, and also that a spectro is able to develop profiles under a wider range of conditions and more consistently than a scanner. I do believe, however, that in the majority of cases, a scanner can be used to derive a printer profile and if proper care and attention is given to the scanner-based profiling methods, the resulting profiles will be nearly as good as those generated from a spectro.

No comment about being platform agnostic?

Just one: use Virtual PC. ;-) You only have to use it to build the profile. Once built, the profiles will work under Mac OSX or anything else.

--
Mike
http://www.ddisoftware.com

 

[fotografer]

Hi Mike,

Thanks for making things so clear in your response to Andy. Indeed the Profile Prism is the poor man's tool of getting good enough profiles for printers and I think it does a marvellous job even with my lowly cp1700. The concept of using two targets in the same scan and profiling the scanner in that same scan and then map the printer scan into LAB is a very smart design.

This reminds me a little of how sometimes professionals view 'affordable' (but to a non-professionals - an excellent product for the price) profiling packaging as not up to scratch. Sure by doing two scans one for profiling the scanner and then to convert and map the other for the printer may not be ideal for all cases, but what we are talking about here are improved profiles that are way better than the canned ones supplied by the printer manufacturers.

But I have to agree with Andy with one point, it seems that PP (at least for my lowly scanner and printer) does seem to profile blue to be more 'violet'. I say this with looking at the PDI-target. Perhaps I am used to my uncallibrated monitor version of 'blue', or perhaps blue is indeed the most difficult to profile.

Mind you, I have not gone beyond the 'automatic' calibration (simply print, leave for 24 hours, and then scan both target, get the corners right and let the software profile away!). Perhaps there are finer points of tweaking colors?

Also, it seems the profiles generated gave more 'grainy' feel to the unprofiled ones, but instances I use Qimage to print. Is it because this is caused by the Lanczos (spelling?) interpolation and the smart sharpening? I have tried reducing the sharpening the details became lost with little reduction in grainy pictures...

I am sure this question has been asked before, but I have missed it: Do you also perhaps plan to come up with scan target that does only gray-scale profiling? Do you think the limitation of using scanner as a poor man's make-shift spectrometer will not yield neutral-enough grays with PP?

And the those Mac users out there, indeed why not try it with virtual PC?

Just one: use Virtual PC. ;-) You only have to use it to build
the profile. Once built, the profiles will work under Mac OSX or
anything else.

--
Fotografer
...like, a total himbo

 

[Andrew Rodney]

No comment about being platform agnostic?

Just one: use Virtual PC. ;-) You only have to use it to build
the profile. Once built, the profiles will work under Mac OSX or
anything else.

We Mac users are cursed that we can run both . I don't have or use Virtual PC. That's too scary for me . In fact I haven't even booted into OS9 in months. Sure you don't want to start messing with UNIX?

Thanks for all the information, very interesting! And I do recall that Wizwiwy did have a two part target (you got a target that they printed on a LightJet and a TDF file) and you taped down the print portion and scanned both.
--
Andrew Rodney
http://www.digitaldog.net

 

[Phill Brannon]

Meanwhile back at the ranch, did Larry Beemer ever get a profile set up??

 

[J A K]

And I do recall
that Wizwiwy did have a two part target (you got a target that they
printed on a LightJet and a TDF file) and you taped down the print
portion and scanned both.

Hi Andrew:

Wiziwyg (or however it is spelled) never worked because (1) the printer target print came out too dark and (2) the scanned information was at the mercy of the luminance uniformity (or lack of it) of the scanner. There was a third problem - even if you had a perfect scanner, the computed profile did not have a smooth tonal range (e.g. you could see Mach banding in the sky as you went up from the horizon). In my opinion, it was hopeless from the start.

The PP reflectance standard has a gray band around each of about 750 (or so) colors; this allows the software to compensate for the scanner's luminance nonuniformity. Mike Chaney does not come right out and say this in the "how to profile your printer" instructions; however, it is STRONGLY implied in the "how to profile your camera" instructions. He clearly states this is how he compensates for image nonuniformity due to light falloff associated with camera optics. IMO, the level of compensation afforded by the gray band is what makes the PP system work so well.

You are well aware that the GretagMacbeth D60 target for profiling cameras is based on a somewhat similar idea (e.g. the D60 color chart has a repeating pattern around the perimeter and the center).

Regards,

Joe Kurkjian, Pbase Supporter

http://www.pbase.com/jkurkjia

 

[Larry Beemer]

My attempts at profiling my printer are at a stand still at the moment. I discovered that the monitor profile(s) I've been trying to create are bad due to perhaps a bad Spyder. Getting a serious yellow/green cast on images opened in Photoshop.....but that's a whole 'nother can o' worms. Heck, even a basic Adobe Gamma monitor profile isn't working now....and it used to!! Don't get me started.........

I'm going to have to wait until I get that issue resolved, especially with Pantone Colorvision, before I can continue with the printer profiling.

I will come back here and post when I get things back in order and I'm able to attempt the printer profiling again. Unfortunately, it may be a while.

In the mean time....anybody wanna buy a comb? I don't think I'll be needing it anymore.

LB

 

[Larry Beemer]

Hey Paul!! Fancy meeting you here.

Regarding my picture in the calendar....too bad it's being replaced with the "new and improved" version. Ironically, I knew this would happen all along so it's really no surprise. However, you can BET I'll have the "old version" up on my wall come January! LOL

LB

 

[pnthobby]

Larry,

I'll keep yours up here in my office. I have ordered the new calendar, but I'm sure the "new January" will not be any better than yours! I didn't even realize it was there 'til I switched to December and realized there was another page back there!!! Hope to see you later this season... Happy holidays, and seasons greetings to you and yours....

Paul Weber
--
'No matter where you go, there you are...' Buckaroo Banzai