Pages: 1 ... 14 15 [16] 17 18 ... 78   Go Down

Author Topic: DCamProf - a new camera profiling tool  (Read 767334 times)

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: DCamProf - a new camera profiling tool
« Reply #300 on: May 30, 2015, 12:26:08 pm »

the darker patch quality
but see the note about measuring those patches - yes, you improve raw RGB data - but where you are going to get the measurements of the same accuracy ? unless you are not going to use those patches in the first place or you have a magnitude better measuring device for your target ... it is interesting to know what are the actual limits and how you can actually find out that the patch that you are trying to measure with i1pro2 or colormunki is not the one that you can reliable measure with them... does anybody has any numbers ?
Logged

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: DCamProf - a new camera profiling tool
« Reply #301 on: May 30, 2015, 12:28:28 pm »

> darker patches different chromacity wise from lighter patches
Different chromaticity does not matter with a target composed of 5 pigments only.
you mean target that is printed on a printer with inks, right ? and the higher end targets simply do not (except black patches) patches that are not reliably measured with i1pro2-level instruments by design ?
Logged

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
Re: DCamProf - a new camera profiling tool
« Reply #302 on: May 30, 2015, 12:30:02 pm »

No, I mean ColorChecker series. There were times when 6 pigments were in use, but one of those got into hazardous materials list.
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3267
Re: DCamProf - a new camera profiling tool
« Reply #303 on: May 30, 2015, 04:57:17 pm »

A good way to test the performance of a test target is to measure it, make a profile with camera SSFs and then test the resulting profile against real spectral data from a database. I've done a basic test of that type in a neighboring thread.

The X-Rite targets do well, although profiling directly against the desired database you want to match is best of course. The profiling problem is sort of an impossible equation though, if you correct for one set of spectra it may produce contraproductive corrections for another set of spectra. I've seen this when working with SSFs and profiling directly against real spectra. It's not a big problem though, but the point is that one needs to make some sort of average correction for typical spectra, which means that the search for the perfect target is futile. It's not certain that a profile made against real spectra will work better than a profile made against an artifical target.

The number of pigments matters, but I guess also the spectral shape of the pigments. The X-rite CC24 have nice shapes of their spectra (you can plot it to see) probably because the pigments are broad band and more overlapping. Compared to patches printed on my inkjet which have more narrow band pigments that can achieve higher saturation, but end up with less varied spectra anyway. All that is an area worth more investigation though, from my point of view.
« Last Edit: May 30, 2015, 05:06:55 pm by torger »
Logged

GWGill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 608
  • Author of ArgyllCMS & ArgyllPRO ColorMeter
    • ArgyllCMS
Re: DCamProf - a new camera profiling tool
« Reply #304 on: May 30, 2015, 10:36:36 pm »

txt2ti3 when fed with CGATS file with raw RGB data and .ti3 from chartread with spectral data there will not include spectral details in the output, unless I replace "SPEC_" in source .ti3 with "nm", "NM_", "SPECTRAL_NM_",  "R_" or "SPECTRAL_"... it is certainly for a good reason, but not convenient !
txt2ti3 is intended to input foreign file formats, not .tiX files.  What are you trying to achieve ? - there may be some other way of doing it that is more appropriate.
Logged

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: DCamProf - a new camera profiling tool
« Reply #305 on: May 31, 2015, 12:51:12 am »

What are you trying to achieve ? - there may be some other way of doing it that is more appropriate.

for example I have a cgats file with raw rgb data

case #1, I have spectral measurements produced by something else in cgats format
case #2, I have spectral measurements produced by argyll chartread (.ti3 has also XYZ and/or LAB, even I do not need to have them, no option to exclude... not a bid deal, but for consistency again it is good to be able to say that I only need spectral data)

case #1, using txt2ti3 to combine the files gives the output that I like : raw rgb from the actual shot + spectral data from the target measurements, leaving whatever software further down the chain to use the spectral data as it can/need
case #2, using txt2ti3 to combine the files gives me raw rgb + but no spectral data included, which kind of makes using chartread more difficult to use in this situation

inconsistent at least ! what I 'd love to have is to use argyll chartread to measure a target and then use txt2ti3 to combine raw rgb data with just spectral data from chartread output... w/o using text editors, etc... yes, I understand the origin how txt2ti3 was intended to be used, but

PS: and if the target measuments were not in 10nm (but for example like -H mode in 3.33x increments), why it is necessary to strip intermediate measurements from the source file and output only SPEC_380, SPEC_390, etc ?
« Last Edit: May 31, 2015, 12:56:54 am by AlterEgo »
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3267
Re: DCamProf - a new camera profiling tool
« Reply #306 on: May 31, 2015, 06:47:49 am »

Been reading more on CATs. It seems like all CATs, including the linear Bradford, are derived from corresponding color data sets and thus try to model "color inconstancy", that is the subtle color apperance differences that still exist under various light.

There is no established model specifically designed for "relighting", that is predict XYZ coordinates for a different illuminant that gives the same result as re-intergrating the reflectance spectra with the new illuminant, although there are papers describing methods for doing that, such as this: http://www.cs.sfu.ca/~funt/2014_CIC22_MIRZAEI_GaussianPrediction.pdf and those methods make more accurate results than the established CATs. However, as a DCamProf workflow should include spectra in the normal case implementing this type of transform will not be a top priority. 0.7.1 will use Bradford rather than CAT02 when spectral data is missing for relighting, as the test data I've seen in papers show that it performs better for that particular task, but as said it's still not designed for it and there are better models for it.

The other more interesting question I'm trying to find an answer to is how well the CAT02 keeps appearance over illuminant change. The problem DCamProf faces is that profile must be made relative to D50 (the profile connection space), but the profile may have say StdA as calibration illuminant and we may want to keep those color appearances, that is convert to corresponding colors in D50 rather than relight the reflectance spectra. It's very clear that CATs are highly approximate, they will introduce errors. Therefore I wanted to find a comparison with a spectral model of perfect color constancy, that is relighting from spectra. The only paper I've found on this so far is this:
http://www.rit-mcsl.org/fairchild/PDFs/PRO28.pdf
But unfortunately the experimental data is too thin to make any conclusions, it does indicate though that the spectral color constant model might not have larger errors than CAT02 will have. It will indeed be in error by principle, show constant color when we know there should not be constant color, but the actual average DeltaE might be lower than for CAT02.

So I still don't know if the current default (use spectral color constant model) is the right one, or if I should change to CAT02. The investigation continues...
« Last Edit: May 31, 2015, 06:53:08 am by torger »
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: DCamProf - a new camera profiling tool
« Reply #307 on: May 31, 2015, 08:17:36 am »

but see the note about measuring those patches - yes, you improve raw RGB data - but where you are going to get the measurements of the same accuracy ?

Just like the capture result will get improved precision from averaging, so will the target measurement improve from averaging multiple measurements. The read/electronic noise, which is a larger percentage of dark patch measurements, will reduce by the square root of the number of reads. So averaging 4 reads, will reduce the noise to 50% of a single read, 16 reads will reduce he noise to 25%. How many is enough? That depends on the level of noise and if it produces erratic readings that influence the precision notably.

Cheers,
Bart
« Last Edit: May 31, 2015, 08:23:18 am by BartvanderWolf »
Logged
== If you do what you did, you'll get what you got. ==

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: DCamProf - a new camera profiling tool
« Reply #308 on: May 31, 2015, 10:22:31 am »

Just like the capture result will get improved precision from averaging, so will the target measurement improve from averaging multiple measurements. The read/electronic noise, which is a larger percentage of dark patch measurements, will reduce by the square root of the number of reads. So averaging 4 reads, will reduce the noise to 50% of a single read, 16 reads will reduce he noise to 25%. How many is enough? That depends on the level of noise and if it produces erratic readings that influence the precision notably.

Cheers,
Bart

I have colorchecker sg, I am measuring it with i1pro2, I am getting black patches (those along the edge) as L* = 10.x-11.x, I check the data from BabelColor Patchtool (I'd assume 1:1 from ProfileMaker) and from Iliah Borg (from makeinputicc) and I see that those patches there are L* = 6.x-7.x

Do you think that is an error that can be eliminated by many measurements ? because all my black patches like this and that is = 14 measurements, same L*'s... so do I have a different copy of CCSG (some change in manufacturing process, it differs visually from for example the one that you can see in I-R shots - there the name is on top in the middle, my has the name on top, but on the left side and "Digital SG" is black letters with white background) or does i1pro2 simply have an issue with measuring black patches that can't be eliminated by multiple measurements or something else (what) ? lighter patches way better matching L* wise between my measurements and BabelColor|Iliah.

PS: when I am measuring black trap with i1Pro2, I am getting L* reading from perfect 0.0000 to 0.1 ... so error (including operator's too) is there, but that error does not seem to be 1-s of L*
« Last Edit: May 31, 2015, 11:35:40 am by AlterEgo »
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3267
Re: DCamProf - a new camera profiling tool
« Reply #309 on: May 31, 2015, 10:37:09 am »

Concerning relighting transform; I've compared Bradford and CAT02 with DCamProf's ability to render an artificial spectra from an XYZ coordinate, and then integrate with the new illuminant, the results are that Bradford and CAT02 has very similar results, and going through an artifical spectra is almost always more accurate, often sub-1 Delta E, while the CATs is often up at 2-4. This algorithm is similar to what is described in the paper linked earlier, although much more computationally expensive.

To the next version the default recommendation for relighting transform will be via artificial spectra, the drawback is that it's very slow though. It's still always best to have measured spectra of your target of course, but if you don't have a spectrometer and only have XYZ coordinates in your reference file and you want to calibrate for a different illuminant you need a relighting transform.

I haven't found any standard name for "relighting transform", some call it "color signal prediction", but in most contexts people say just "CAT", which is confusing as all CATs are designed to predict corresponding colors (that is which tristumulus value to use to produce the same appearance), not predict which tristimulus value the same object will have under a new illuminant. Only if you assume that color constancy is 100% perfect they will be the same, and the current established CATs do not.
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: DCamProf - a new camera profiling tool
« Reply #310 on: May 31, 2015, 10:56:47 am »

I have colorchecker sg, I am measuring it with i1pro2, I am getting black patches (those along the edge) as L* = 10.x-11.x, I check the data from BabelColor Patchtool (I'd assume 1:1 from ProfileMaker) and from Iliah Borg (from makeinputicc) and I see that those patches there are L* = 6.x-7.x

Do you think that is an error that can be eliminated by many measurements ? because all my black patches like this and that is = 14 measurements, same L*'s... so do I have a different copy of CCSG (some change in manufacturing process, it differs visually from for example the one that you can see in I-R shots - there the name is on top in the middle, my has the name on top, but on the left side and "Digital SG" is black letters with white background) or does i1pro2 simply have an issue with measuring black patches that can't be eliminated by multiple measurements or something else (what) ? lighter patches way better matching L* wise between my measurements and BabelColor|Iliah.

The differences between multiple samples of the same target will gravitate towards its real mean value (somewhere between 10 and 12 in your case), as the random noise cancels out with each additional sampling of the same patch. But I think that the difference between production runs of the SG plays a role here.

Quote
PS: when I am measuring black trap with i2Pro2, I am getting L* reading from perfect 0.0000 to 0.1 (including operator's too)... so error is there, but that error does not seem to be 1-s of L*

That suggests that there is no dust/dirt or calibration issue in play here, but rather a variation in the SG targets.

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
Re: DCamProf - a new camera profiling tool
« Reply #311 on: May 31, 2015, 10:57:51 am »

>  haven't found any standard name for "relighting transform", some call it "color signal prediction", but in most contexts people say just "CAT"

Dr. Fairchild was using "spectral adaptation". I use sCAT as an abbreviation.
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3267
Re: DCamProf - a new camera profiling tool
« Reply #312 on: May 31, 2015, 11:37:08 am »

sCAT sounds nice, I'll borrow that!
Logged

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
Re: DCamProf - a new camera profiling tool
« Reply #313 on: May 31, 2015, 12:05:14 pm »

Checked 3 SG targets from different manufacturing batches, Spectroscan/Spectrolino vs. i1Pro - it looks like there are no manufacturing variations to speak of. Higher values in case of i1Pro are consistent and seem to be caused by the in-instrument flare.
Logged

Bart_van_der_Wolf

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 8913
Re: DCamProf - a new camera profiling tool
« Reply #314 on: May 31, 2015, 12:13:28 pm »

Checked 3 SG targets from different manufacturing batches, Spectroscan/Spectrolino vs. i1Pro - it looks like there are no manufacturing variations to speak of. Higher values in case of i1Pro are consistent and seem to be caused by the in-instrument flare.

Hi Iliah,

Thanks for checking. I do wonder, measuring a black trap apparently gives virtually no signal, and black should not reflect much, I would not expect enough to cause such an amount of internal glare. If a lighter patch is measured in the same acceptance angle of the instrument yes, but with only 'black' in view?

Cheers,
Bart
Logged
== If you do what you did, you'll get what you got. ==

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3267
Re: DCamProf - a new camera profiling tool
« Reply #315 on: May 31, 2015, 12:26:57 pm »

sCAT sounds nice, I'll borrow that!

Oh, realized it's not the same thing. Fairchild's "spectral adaptation" is CAT based on full spectral data, the human adaptation is still modeled just like in an ordinary CAT. In the paper he uses a "color constant" version as reference though, which is the exact same thing as a relighting transform.
Logged

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
Re: DCamProf - a new camera profiling tool
« Reply #316 on: May 31, 2015, 01:01:40 pm »

Let's consider a workflow which was at some point SOP in (at least) one of museums for reproduction work:
- shoot a target at studio light
- measure light spectrum
- re-calculate the target reference for measured spectrum
- convert "light-adjusted" target reference to XYZ
- calculate a matrix to convert "device RGB" to "light-adjusted" XYZ
- use CAT to D50/D65/Ill.A to get matrices needed for a dcp profile

You can easily see there are serious problems with this workflow, and also they are converting to XYZ at a premature stage.

sCAT is just a term, like log (logarithm). It needs qualification as to the method used. What sCAT emphasizes is that the adaptation is based on spectra directly, not just on XYZ convolution.
Logged

Iliah

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 770
Re: DCamProf - a new camera profiling tool
« Reply #317 on: May 31, 2015, 01:18:25 pm »

Error kicks in below L*=20, which, in the words of one of X-Rite guys on a seminar, is as deep as one may want to go for _printer_ profiling.
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3267
Re: DCamProf - a new camera profiling tool
« Reply #318 on: May 31, 2015, 01:43:16 pm »

Very interesting on the spectrometer performance on dark patches.

DCamProf is not ideal for copy profiles as it only makes a 2.5D LUT, that is a specific chromaticity gets the same correction regardless of lightness. With say an IT-8 photographic target for copying photographs printed in the same media a 3D LUT will probably achieve a tiny bit better result.

I chose to make 2.5D profiles as I think it's better for generic profiles where exposure will vary as well as object spectra.

This means that one does not really need dark patches in the test target, except for highly saturated colors close to the line of purples that must be dark otherwise the saturation cannot be reached. (There's also a possibility that dark patches can add some spectral variation in the target by mixing in more colorants.)

From DCamProf's perspective many of the "skin tone" patches on the Colorchecker SG are redundant as they are (nearly) just darker versions of a lighter patch, and indeed the SG does not make more accurate skintone profiles than the 24 patch colorchecker when I compare with the lippmann2000 data.
« Last Edit: May 31, 2015, 01:47:15 pm by torger »
Logged

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: DCamProf - a new camera profiling tool
« Reply #319 on: May 31, 2015, 02:26:09 pm »

the 24 patch colorchecker when I compare with the lippmann2000 data.

consider 2 cases :

1) -p cc24 -p skin-white.ti3

2) -p skin-white.ti3 -p cc24


case #1 = 2 patches from CC24 will exclude all the patches in skin-white.ti3 (resulting target = CC24 patches, 24 of them) and case #2 = all patches in skin-white.ti3 will exclude only those 2 patches from CC24 (resulting target = all 78 patches from skin-white.ti3 and 24-2 patches from CC24 = 100 patches)

so is case #2 going to be really (noticeably) better for raw with caucasian skin tone in your opinion ?

Logged
Pages: 1 ... 14 15 [16] 17 18 ... 78   Go Up