Topic: will X-Rite wake up and smell the coffee? (Read 69273 times)

digitaldog · « **Reply #80 on:** July 25, 2011, 11:35:59 am »

Quote from: Mark D Segal on July 25, 2011, 10:48:07 am

After making a profile with Spectraview and testing it with Patchtool I was not completely satisfied so I made another one with BasicColor (using the NEC colorimeter), ran that one through Patchtool and the dE values were lower.

So without even asking about the specifics of the patches measured, the lower dE is solely due to the profile type, that’s your analysis? Both software products sent exactly the same color values to the instrument and used the same math to control the panel, the sole difference here in the lower dE is the profile type? You sure?

And your explanation as to why NEC is incapable or refuses to build a LUT profile is?

Mark D Segal · « **Reply #81 on:** July 25, 2011, 12:58:19 pm »

Quote from: digitaldog on July 25, 2011, 11:35:59 am

So without even asking about the specifics of the patches measured, the lower dE is solely due to the profile type, that’s your analysis? Both software products sent exactly the same color values to the instrument and used the same math to control the panel, the sole difference here in the lower dE is the profile type? You sure?

And your explanation as to why NEC is incapable or refuses to build a LUT profile is?

Yes of course - if you use the same patches - a good representative sample - different from the verification routine of the profiling software itself so we aren't involved in self-fulfilling prophecies - for testing both profiles and the dE values come out differently, it is fair to conclude on which profile is better. This is a repeatable, scientific procedure. I don't know what you mean by the "specifics of the patches being measured" - you should ask Danny Pascale about that. Patchtool has a number of bespoke patch sets specifically prepared for these purposes. And yes, as for the results, I can read a dE number thrown-up by the results panel of the testing application. I can even read two such numbers and compare them. So am I SURE? - yes I am SURE.

I have discussed the issue extensively with NEC. Their response is that their PA series displays are linear and linear displays don't need LUT profiles - the nine points of a matrix profile are perfectly sufficient. If their premise were correct their argument would be correct.

digitaldog · « **Reply #82 on:** July 25, 2011, 01:02:59 pm »

Quote from: Mark D Segal on July 25, 2011, 12:58:19 pm

Quote
Yes of course - if you use the same patches - a good representative sample - different from the verification routine of the profiling software itself so we aren't involved in self-fulfilling prophecies - for testing both profiles and the dE values come out differently, it is fair to conclude on which profile is better.

Hold on skipper. You used NEC and BasICColor right? You presumably set the same targets for calibration right? How else did you build the two profiles? Now you KNOW that both products sent the same number of color patches to the display using the same colors within color space to build the profile? That the only differences here in terms of your ‘results’ are one profile was LUT and the other was Matrix?

Quote
I have discussed the issue extensively with NEC. Their response is that their PA series displays are linear and linear displays don't need LUT profiles - the nine points of a matrix profile are perfectly sufficient. If their premise were correct their argument would be correct.

So how is their premise wrong?

Mark D Segal · « **Reply #83 on:** July 25, 2011, 01:13:55 pm »

Quote from: digitaldog on July 25, 2011, 01:02:59 pm

Hold on skipper. You used NEC and BasICColor right? You presumably set the same targets for calibration right? How else did you build the two profiles? Now you KNOW that both products sent the same number of color patches to the display using the same colors within color space to build the profile? That the only differences here in terms of your ‘results’ are one profile was LUT and the other was Matrix?

So how is their premise wrong?

I'm not "skipper".

No - each application has its own patch set for building the profiles. I used the same patch set in PatchTool for testing the profiles. So if you want to argue that BasicColor is a better application for building profiles than Spectraview for reasons other than the difference in profile type, that proposition can be tested by asking BasicColor to make a matrix profile (if my memory serves me correctly I believe it can - I'm not near the software to check this) followed by a LUT profile and testing the pair through PatchTool. I'm not in a position to do that just now, but early next month I will be. If I have time, I'll try it and let you know the outcome. But I think it stands to reason that when dealing with non-linear situations, a profile calculated from several hundred data points is highly likely to be more accurate than one calculated from nine data points.

The linearity premise is incorrect.

digitaldog · « **Reply #84 on:** July 25, 2011, 01:29:26 pm »

Quote from: Mark D Segal on July 25, 2011, 01:13:55 pm

No - each application has its own patch set for building the profiles.

So that had no effect on the profile?

Quote

So if you want to argue that BasicColor is a better application for building profiles than Spectraview for reasons other than the difference in profile type, that proposition can be tested by asking BasicColor to make a matrix profile

I’m not as yet arguing anything other than you analysis that two profiles tested by a third product (Patchtool) resulting in a lower dE is solely due to the differences between LUT and Matrix!

Even if you build a Matrix profile in BasICColor and it produces a ‘better’ report in PatchTool, that would not prove that the better results from the LUT profile were due to it being a LUT (the test would actually imply the opposite). It might prove that with the patches sent to the display to build either or both profiles, or the math, or how the product controls the panel is better. And that’s been my point since day 1! No one has proven as yet that a LUT profile is better than a Matrix profile all things being equal (which for good science all things DO need to be equal and you’ve got too many variables to say with any certainty its the profile structure).

Quote

But I think it stands to reason that when dealing with non-linear situations, a profile calculated from several hundred data points is highly likely to be more accurate than one calculated from nine data points.The linearity premise is incorrect.

IT doesn’t stand to reason. Because you haven’t provided the proper testing to make it a reasonable analysis of the process. I’m not sure you can. Each product provides too many differences in how it calibrates and then builds a profile from the data it sends to the instrument that I seriously doubt you can prove that the lower dE is solely due to the structure of a profile. But yet, that’s been your take a for a long time.

IF you want to say, “I’ve used as similar settings between two dissimilar software products to calibrate and build a profile, and the profile from product A in PatchTool provides a lower dE report” then fine. To suggest the reason is due to the structure of the profile is unproven. Based on the number of different software products I’ve tested over the years, on the same display using the same instrument, its not at all surprising to find differences in quality (assuming you believe a test like that in PatchTool is the sole criteria of quality). Saying its because the display faced north instead of south is just silly stuff. Saying the results are better because one profile was LUT and the other was Matrix is as silly until all other variables that affect the process are eliminated and you’re far from that.

Next question. You presumably set both software products for the same target calibration targets? Did they produce identical results (measured or otherwise)? I’d think not. Is one ‘more right’ than the other?

terrywyse · « **Reply #85 on:** July 25, 2011, 02:14:08 pm »

Quote from: digitaldog on July 25, 2011, 01:29:26 pm

Even if you build a Matrix profile in BasICColor and it produces a ‘better’ report in PatchTool, that would not prove that the better results from the LUT profile were due to it being a LUT (the test would actually imply the opposite). It might prove that with the patches sent to the display to build either or both profiles, or the math, or how the product controls the panel is better. And that’s been my point since day 1! No one has proven as yet that a LUT profile is better than a Matrix profile all things being equal (which for good science all things DO need to be equal and you’ve got too many variables to say with any certainty its the profile structure).

"All things being equal"....isn't this a bit of a straw man argument? It's highly likely that a different patch set will be used for a matrix vs. LUT profile....and both profiles would CERTAINLY use different math during construction of the profile....so you could never have a situation where "all things are equal" so your argument would be rather self-fulfilling. I'm sure you're aware that it would be extremely difficult to get two different display applications to use the same device value patch set....I personally only know of a few that can be coaxed to do that..even then, it's a bit of an under-the-hood/hack kind of exercise.

If the discussion was framed more in terms of....basiCColor Display builds a better LUT profile compared to NEC's matrix profile....then Mark's analysis is valid. Even better....if basICColor built a superior matrix profile compared to NEC's matrix profile, at the very least you could say that NEC's own software for their displays is inadequate.

I'd prefer to think it terms of....."all things being equal, a LUT profile will be no WORSE than a matrix profile...but a matrix profile is MORE likely to be worse than a LUT profile...all things being equal of course"....or something like that.

Regards,
Terry

digitaldog · « **Reply #86 on:** July 25, 2011, 02:36:32 pm »

Quote from: terrywyse on July 25, 2011, 02:14:08 pm

"All things being equal"....isn't this a bit of a straw man argument?

Only for someone who has a theory he can’t prove. If the sole reason the LUT profile is superior is because its a LUT profile, proof its not other measured data, math etc in the process creating the profile is required.

Quote

It's highly likely that a different patch set will be used for a matrix vs. LUT profile....

You think that taking a smaller number of patches spread in a narrower area of color space would not have an effect over a product that sampled either a larger sample or a sample of colors in a wider area of color space? Its the LUT, its not the patches, that’s Marks assumption.

Quote

and both profiles would CERTAINLY use different math during construction of the profile....

Can’t be the patches or the math, its the structure of the profile. That’s Mark’s take.

Quote

so you could never have a situation where "all things are equal" so your argument would be rather self-fulfilling. I

No, it just makes Marks assumptions about the profile far more difficult to prove. One questions that with all these differences, how one came to the conclusion its solely the profile structure. It makes accepting the premise so much more difficult. And yet, its been said its not a flat earth theory. Yet its Mark’s theory to explain and prove. I don’t have to disprove it, that’s not how the science works. Mark have to prove the theory, now you are saying that difficult to do. I don’t disagree. Does’t change the facts (or lack thereof).

Quote

I'm sure you're aware that it would be extremely difficult to get two different display applications to use the same device value patch set....

I know people who could write the software to do this. Maybe you can’t. I can’t. But then again, there is this theory that keeps showing up here but there’s nothing to back it up.

Quote

If the discussion was framed more in terms of....basiCColor Display builds a better LUT profile compared to NEC's matrix profile....then Mark's analysis is valid.

That’s an easier theory to accept at this point. Personal experience suggests that is very possible. At least if we agree that sending X number of as yet, undefined patches through two profiles and analyzed by PatchTool is the only criteria of ‘a better profile’.

Quote

Even better....if basICColor built a superior matrix profile compared to NEC's matrix profile, at the very least you could say that NEC's own software for their displays is inadequate.

Well inadequate in terms of a dE value of as yet undefined colors in an undefined area of color space. But that is simply one data point. It tells us nothing about other aspects of calibration and profiling a display. For example, if I could produce a lower dE of defined patches at the expense of purity in other areas of the display from center (where the data is measured), would that be adequate? Or if I asked for a 300:1 contrast ratio but in order to get a lower dE of said patches, the actual calibration produced 400:1, would that be adequate?

The idea that a lower dE test points to the profile structure alone is so far nonsensical based on the process used to come to that conclusion. Now one has to ask, if the dE of product #1 is 2.9 and the other is 3.5, does that automatically make product #2 inadequate? Maybe.

Quote

I'd prefer to think it terms of....."all things being equal, a LUT profile will be no WORSE than a matrix profile...

The key part of the sentence is what isn’t happening in the analysis, all things being equal.

Alan Goldhammer · « **Reply #87 on:** July 25, 2011, 02:45:51 pm »

If one wanted to test this out, ArgyllCMS allows you to generate matrix and LUT profiles from the same patch set. I have prepared matrix profiles using both Spectraview and Argyll using the NEC Puck that comes with Spectraview. Obviously Spectraview is much easier to use but Argyll allows for greater configurability of the monitor. I'm of a mixed view on this since the time and effort needed to get Argyll working is considerable (and we had a discussion of this a month or so ago on the thread Ethan Hansen started). I've not done any kind of measuring to see which profile is "better" but that's something I do intend to look at.

digitaldog · « **Reply #88 on:** July 25, 2011, 02:49:17 pm »

Quote from: Alan Goldhammer on July 25, 2011, 02:45:51 pm

If one wanted to test this out, ArgyllCMS allows you to generate matrix and LUT profiles from the same patch set.

That would be a very interesting test, but would provide no further evidence that NEC produces an inferior result than BasICColor (in terms of the one dE analysis) because NEC doesn’t write a LUT profile. It would tell us scientifically how LUT vs. Matrix profiles preform with ArgyllCMS.

Alan Goldhammer · « **Reply #89 on:** July 25, 2011, 02:58:01 pm »

Quote from: digitaldog on July 25, 2011, 02:49:17 pm

That would be a very interesting test, but would provide no further evidence that NEC produces an inferior result than BasICColor (in terms of the one dE analysis) because NEC doesn’t write a LUT profile. It would tell us scientifically how LUT vs. Matrix profiles preform with ArgyllCMS.

Exactly right. The problem with any of the preceding discussion is the number of variables that would need to be controlled. I've done enough lab work in my time to know when to hold them and when to fold them. This is a case of the latter as once can get so preoccupied with numbers that you forget to go out and take pictures!

Mark D Segal · « **Reply #90 on:** July 25, 2011, 09:36:36 pm »

Andrew, I agree with you that to be dead-sure of a hypothesis one should neutralize all variables but one and test for that one. We've seen in this discussion that this would be very difficult to do. The furthest I can take it, when I'm in a position to do so, is to make both matrix and LUT profiles from the same software and measure the results. But if the measurement approach itself is not trustworthy, then there is no concrete answer to this issue and no point starting - again. But I'd be surprised. To avoid wasting my time, I'll do some further consulting on what BasicColor does and what PathcTool does before proceeding, Finally, I wonder why these LUT profiles were developed in the first place if these simple matrix profiles are no worse?

Mark D Segal · « **Reply #91 on:** July 25, 2011, 09:38:12 pm »

Quote from: digitaldog on July 25, 2011, 02:49:17 pm

That would be a very interesting test, but would provide no further evidence that NEC produces an inferior result than BasICColor (in terms of the one dE analysis) because NEC doesn’t write a LUT profile. It would tell us scientifically how LUT vs. Matrix profiles preform with ArgyllCMS.

That's correct, but it does nail as many variables as possible and therefore provides additional useful insight, so I would encourage Alan to do this when he has the time and inclination!

digitaldog · « **Reply #92 on:** July 25, 2011, 09:47:27 pm »

Quote from: Mark D Segal on July 25, 2011, 09:36:36 pm

Finally, I wonder why these LUT profiles were developed in the first place if these simple matrix profiles are no worse?

For one, basICColor is a product that is for sale for any display system. There are such systems that will benefit from such a profile. But not a display like a SpectraView or Eizo. Think a POS display purchased because it cost a buck ninety five.

Ernst Dinkla · « **Reply #93 on:** July 26, 2011, 03:50:49 am »

Graeme Gill of ArgyllCMS - Dispcal has written some good observations on the differences between matrix and LUT profiling for displays and I do not think one should try to get more insight than already written down by Graeme.

Quote:
There are two basic choices of profile type for a display, a shaper/matrix profile, or a LUT based profile. They have different tradeoffs. A shaper/matrix profile will work well on a well behaved display, that is one that behaves in an additive color manner, will give very smooth looking results, and needs fewer test points to create. A LUT based profile on the other hand, will model any display behaviour more accurately, and can accommodate gamut mapping and different intent tables. Often it can show some unevenness and contouring in the results though.
End of quote.

http://www.argyllcms.com/doc/Scenarios.html
and more extensive:
http://hoech.net/dispcalGUI/

The reason to include both LUT and Matrix in one profile (BTW display profiles can also carry the calibration data) is that applications differ on which one they use in their color management. To check what the CM uses you got an odd profiling choice in Dispcal where the red and green data are reversed in the matrix side of the LUT + matrix profile. So there is a minefield of variables out there if you want to compare the two methods.

met vriendelijke groeten, Ernst

New: Spectral plots of +250 inkjet papers:

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm

The View · « **Reply #94 on:** July 27, 2011, 12:30:55 am »

Quote from: digitaldog on July 25, 2011, 09:47:27 pm

For one, basICColor is a product that is for sale for any display system. There are such systems that will benefit from such a profile. But not a display like a SpectraView or Eizo. Think a POS display purchased because it cost a buck ninety five.

From an email by Basiccolor I know that the Spectraview II software is identical with BasicColor's software (with a few minor adjustments).

digitaldog · « **Reply #95 on:** July 27, 2011, 09:59:26 am »

Quote from: The View on July 27, 2011, 12:30:55 am

From an email by Basiccolor I know that the Spectraview II software is identical with BasicColor's software (with a few minor adjustments).

Considering they are coded and designed by completely different entities, I find that difficult to believe.

Mark D Segal · « **Reply #96 on:** July 27, 2011, 11:07:23 am »

Quote from: digitaldog on July 27, 2011, 09:59:26 am

Considering they are coded and designed by completely different entities, I find that difficult to believe.

I agree.

BasicColor provides a version of its software for the European PA Series, and that may be the meaning of the communication.

Curt Peters · « **Reply #97 on:** August 19, 2011, 02:07:46 pm »

UPDATE: XRITE has listened and has introduced a rebate program. See it here: http://www.xritephoto.com/custom_page.aspx?PageID=271 I'm really happy that XRITE has done this.... I believe it's a win win for all.
Thank You XRITE..... the coffee smells real good now!

Ronny Nilsen · « **Reply #98 on:** August 19, 2011, 03:12:34 pm »

Quote from: Curt Peters on August 19, 2011, 02:07:46 pm

UPDATE: XRITE has listened and has introduced a rebate program. See it here: http://www.xritephoto.com/custom_page.aspx?PageID=271 I'm really happy that XRITE has done this.... I believe it's a win win for all.
Thank You XRITE..... the coffee smells real good now!

But it looks like the rebate is only good for US and Canada.

Not any good on my side of the planet.

Ronny

John MacLean · « **Reply #99 on:** August 19, 2011, 10:16:02 pm »

Too bad X-Rite isn't offering an instant rebate. If your snail mail gets lost, then X-Rite wins!

It seems like B&H isn't even aware there is one:
http://www.bhphotovideo.com/c/product/759155-REG/X_Rite_EOPROF_UPGA_i1Publish_Software_Upgrade_A.html

Author Topic: will X-Rite wake up and smell the coffee? (Read 69273 times)