Pages: 1 ... 10 11 [12] 13   Go Down

Author Topic: New profile making software: Lumariver Profile Designer  (Read 27292 times)

AlterEgo

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1995
Re: New profile making software: Lumariver Profile Designer
« Reply #220 on: October 12, 2017, 01:24:26 PM »

It "works" with caveats.

works (not "works" !) with caveats that you need some measured (even not ideally) reference data from other cameras, yes ... for us hoi polloi who are using regular mass produced C&N&S&O&P cameras


Logged

daicehawk

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 71
Re: New profile making software: Lumariver Profile Designer
« Reply #221 on: October 12, 2017, 01:26:43 PM »

I apologize if this is old news - I just received an email regarding v 5.5 of BabelColor's Patch Tool.  The email notes that one of the new features is the "Spectrum Generator"

From: http://www.babelcolor.com/patchtool_spectrum_generator.htm

Any thoughts on the usefulness of this feature as it relates to using spectral data in Profile Editor?

Thanks,

kirk
One to many relationship. one XYZ can be a gazzillion of different spectra. I cannot see any use of this.
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 848
Re: New profile making software: Lumariver Profile Designer
« Reply #222 on: October 12, 2017, 02:18:07 PM »

One to many relationship. one XYZ can be a gazzillion of different spectra. I cannot see any use of this.

Sure, and they're all metamers. But if you look at the spectra of existing pigments and combinations and decompose it, they collapse to a reasonable set of principal components that can, in linear combination, be quite a good match in most all cases.

However the XYZ triplet under one illuminant is certainly not very constraining so the synthesized spectrum is going to be fixed from the 3 principal components that best fit for that particular XYZ. One could just as easily publish a database of XYZ -> Spectra.

I can't imagine much application other than modeling metameric shifts under different illuminants when you don't have object spectra available but want to get some idea based on their large color database.

NOTE: I have re-corrected my first comment. Indeed, when only selecting 3 PCA components (single illuminant) the generated spectrums are the same whether or not the full patch spectrums or just the XYZ (or lab which is convertible to XYZ) is used.
« Last Edit: October 12, 2017, 03:24:39 PM by Doug Gray »
Logged

Alexey.Danilchenko

  • Full Member
  • ***
  • Offline Offline
  • Posts: 167
Re: New profile making software: Lumariver Profile Designer
« Reply #223 on: October 12, 2017, 04:18:32 PM »

Link or actual software product which does interface with a Spectrophotometer and does capture spectral data?
Have you actually read what the post and software was about (to which my post replied)?

The post was referring to spectrum generator from XYZ/LAB samples and was asking of using spectral data from that generator in profile editor. I did say it was not real spectral data since the generator does not measure any.

Is that any clearer now?
« Last Edit: October 12, 2017, 04:23:03 PM by Alexey.Danilchenko »
Logged

Alexey.Danilchenko

  • Full Member
  • ***
  • Offline Offline
  • Posts: 167
Re: New profile making software: Lumariver Profile Designer
« Reply #224 on: October 12, 2017, 04:40:09 PM »

works (not "works" !) with caveats that you need some measured (even not ideally) reference data from other cameras, yes ... for us hoi polloi who are using regular mass produced C&N&S&O&P cameras
Not in this sense - PCA will only give you results very similar to the chosen dataset. Anything that is substantially different will not be recovered correctly. It is only useful to analyze similar data
Logged

GWGill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 407
  • Author of ArgyllCMS & ArgyllPRO ColorMeter
    • ArgyllCMS
Re: New profile making software: Lumariver Profile Designer
« Reply #225 on: October 12, 2017, 09:29:23 PM »

One to many relationship. one XYZ can be a gazzillion of different spectra. I cannot see any use of this.
The point is that what a camera interprets is unlikely to be any of "a  gazzillion of different spectra", instead it is likely to be a spectra that is available in the real world. Using this insight helps guide the trade-offs in creating a camera to standard observer transform.

As for PCA as a technique, the impression I get from reading a few papers on the subject is that it is not quite as good as using actual real world spectra. The problem is that reconstructing spectra from PCA doesn't necessarily result in spectra that you are likely to get from the real world. You can get the flavor of this problem from the realization that using unconstrained reconstruction will happily generate spectra with negative values.

[ Another approach I have read about recently that isn't based on real world spectra statistics, constructs the metamer set for each camera color, and uses the perceptual mean of the set volume as the representative standard observer color. One could tweak that approach to include real world statistics by weighting the mean according to real world metamer likelihood. ]
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 848
Re: New profile making software: Lumariver Profile Designer
« Reply #226 on: October 13, 2017, 12:26:51 AM »

The problem is that reconstructing spectra from PCA doesn't necessarily result in spectra that you are likely to get from the real world. You can get the flavor of this problem from the realization that using unconstrained reconstruction will happily generate spectra with negative values.

I was wondering about how patchtool handles unrealizable colors in those circumstances. The choice is to come as close to the XYZ color as possible with a reflective surface from their dataset or return exact XYZ matches but with spectra allowed to be <0 or > 1.0.   So I checked it with Adobe RGB (255,0,0) which is not realizable by existing reflective surfaces. Sure enough, the PCA spectrum goes negative and well over 100% in places.
Logged

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 633
    • Hikes -more than strolls- with my dog
Re: New profile making software: Lumariver Profile Designer
« Reply #227 on: October 13, 2017, 10:53:37 AM »

With PCA as with many other machine learning techniques you are effectively fitting data and doing pattern recognition.  Results can be quite good if the training set is large enough and of good enough quality.  However you get 'plausible' results - not necessarily 'accurate' ones.  It depends what one is after.

Jack
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3253
Re: New profile making software: Lumariver Profile Designer
« Reply #228 on: October 23, 2017, 10:47:59 AM »

DCamProf also has a spectral generator since quite some time if someone wants to experiment.
 https://www.ludd.ltu.se/~torger/dcamprof.html#generated_spectra

It's more of an experimental feature and I didn't add it to Lumariver Profile Designer. What it is typically used for in DCamProf is when you make a profile based on real spectral measurements but you have some gaps, and then you can fill it out with generated spectra. As many have noted you can have infinite variations of a spectrum for one XYZ coordinate, so what DCamProf does is that it has various models using gaussian distributions to make as smooth spectra as possible for a given coordinate, the rationale being that making a profile based on such spectrum will make it more all-around than using peaky spectra. Regarding "impossible" colors the generator will then fail to make a spectrum.
Logged

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 848
Re: New profile making software: Lumariver Profile Designer
« Reply #229 on: October 23, 2017, 05:59:15 PM »

DCamProf also has a spectral generator since quite some time if someone wants to experiment.
 https://www.ludd.ltu.se/~torger/dcamprof.html#generated_spectra

It's more of an experimental feature and I didn't add it to Lumariver Profile Designer. What it is typically used for in DCamProf is when you make a profile based on real spectral measurements but you have some gaps, and then you can fill it out with generated spectra. As many have noted you can have infinite variations of a spectrum for one XYZ coordinate, so what DCamProf does is that it has various models using gaussian distributions to make as smooth spectra as possible for a given coordinate, the rationale being that making a profile based on such spectrum will make it more all-around than using peaky spectra. Regarding "impossible" colors the generator will then fail to make a spectrum.

Interesting Anders.

I've kicked around the notion if using PCA to find the dominant absorption curves of a printer then using that info to optimize the response of a camera's CFA. Possibly by taking to images, one with D50 ish light the other with Ill. A. One could even add various CFLs/LEDs with known spectra. This should provide enough info to generate 5 or more PCs to characterize a CFA reasonably well. And it should work especially well for colorimetric imaging of that printer's prints.  Know of any work done in that area?
Logged

torger

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 3253
Re: New profile making software: Lumariver Profile Designer
« Reply #230 on: October 24, 2017, 10:25:31 AM »

I've kicked around the notion if using PCA to find the dominant absorption curves of a printer then using that info to optimize the response of a camera's CFA. Possibly by taking to images, one with D50 ish light the other with Ill. A. One could even add various CFLs/LEDs with known spectra. This should provide enough info to generate 5 or more PCs to characterize a CFA reasonably well. And it should work especially well for colorimetric imaging of that printer's prints.  Know of any work done in that area?

I'm not sure if I follow, but if you mean a way to figure out what the spectral sensitivity functions are based on a bunch of color checker shots there is some work in that area done yes. I haven't studied it myself, but you can find some here: http://www.gujinwei.org/research/camspec/ and search for "Code for recovering camera spectral sensitivity from a single image" on that page.
Logged

daicehawk

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 71
Re: New profile making software: Lumariver Profile Designer
« Reply #231 on: October 24, 2017, 06:02:55 PM »

I'm not sure if I follow, but if you mean a way to figure out what the spectral sensitivity functions are based on a bunch of color checker shots there is some work in that area done yes. I haven't studied it myself, but you can find some here: http://www.gujinwei.org/research/camspec/ and search for "Code for recovering camera spectral sensitivity from a single image" on that page.
I still stand by a possibility to tweak a matrix from zero like the DNG Profile maker. Then I`d buy. Maybe a separate product\utility.
Logged

markanini

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 57
Re: New profile making software: Lumariver Profile Designer
« Reply #232 on: October 24, 2017, 06:45:57 PM »

I tried a variation of an example provided by Anders on using Lippman2000 cacuasian spectral data as a virtual target for a SSF profile:
dcamprof txt2ti3 -a "caucasian" -s 1 -f 400,700,2 Reflect_AllCaucasian_400_700_2nm.txt caucasian.ti3
dcamprof make-target -c ssf-600D.json -i D55  -p caucasian.ti3 -p cc24 target-caucasian-D55.ti3
dcamprof make-target -c ssf-600D.json -i StdA -p caucasian.ti3 -p cc24 target-caucasian-StdA.ti3

After feeling somewhat content with results using other virtual targets(CC24,CCSG,AdobeRGB-grid,Munsell) and real profile shots, I'm now really happy with skin tones. Even friends that are not technically inclined have commented about pleasant and realistic skin tones. I'm note sure why but the skin tones improved further still by removing all non neutral patches from the CC24 subset. I'm curious what others here make of this.
Logged

Ethan Hansen

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 87
    • Dry Creek Photo
Re: New profile making software: Lumariver Profile Designer
« Reply #233 on: October 25, 2017, 04:06:07 PM »

I'm not sure if I follow, but if you mean a way to figure out what the spectral sensitivity functions are based on a bunch of color checker shots there is some work in that area done yes. I haven't studied it myself, but you can find some here: http://www.gujinwei.org/research/camspec/ and search for "Code for recovering camera spectral sensitivity from a single image" on that page.

I spent all too long 15 or so years ago attempting to create a camera profiling tool using just such a technique. The starting point was a spectrally measured target. For any tristimulus value a camera records there are an infinite number of spectra that can give the pixel reading. If, however, you iterate through enough known colors, the possible camera matching functions converge. I can probably pull up GATF and other presentations I made on the topic if desired.

The profiles I created were colorimetrically accurate to a fault. We used them for product photography under controlled lighting. The problem was that mathematical accuracy did not translate into pleasing visuals. Discontinuities in the profile translated into ugly transitions in images. Franz Herbert at Integrated Color Solutions took a different approach of prioritizing aesthetics over mathematical accuracy. That ended up working far better in the real world.

daicehawk

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 71
Re: New profile making software: Lumariver Profile Designer
« Reply #234 on: October 26, 2017, 09:12:23 AM »

I spent all too long 15 or so years ago attempting to create a camera profiling tool using just such a technique. The starting point was a spectrally measured target. For any tristimulus value a camera records there are an infinite number of spectra that can give the pixel reading. If, however, you iterate through enough known colors, the possible camera matching functions converge. I PLecan probably pull up GATF and other presentations I made on the topic if desired.

The profiles I created were colorimetrically accurate to a fault. We used them for product photography under controlled lighting. The problem was that mathematical accuracy did not translate into pleasing visuals. Discontinuities in the profile translated into ugly transitions in images. Franz Herbert at Integrated Color Solutions took a different approach of prioritizing aesthetics over mathematical accuracy. That ended up working far better in the real world.
I absolutely concur on the visual correctness over DeltaEs.
Any information on what colors are best to profile by would be great.
And what do you mean by controlled light? Xenon, fluorescent or any daylight simulated light sources?
 
Logged

digitaldog

  • Sr. Member
  • ****
  • Online Online
  • Posts: 12505
    • http://digitaldog.net/
Re: New profile making software: Lumariver Profile Designer
« Reply #235 on: October 26, 2017, 10:34:12 AM »

I spent all too long 15 or so years ago attempting to create a camera profiling tool using just such a technique.
The problem was that mathematical accuracy did not translate into pleasing visuals.
We've both spent about the same time coming up with identical conclusions!
Logged
Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 848
Re: New profile making software: Lumariver Profile Designer
« Reply #236 on: October 26, 2017, 02:53:02 PM »

I spent all too long 15 or so years ago attempting to create a camera profiling tool using just such a technique. The starting point was a spectrally measured target. For any tristimulus value a camera records there are an infinite number of spectra that can give the pixel reading. If, however, you iterate through enough known colors, the possible camera matching functions converge. I can probably pull up GATF and other presentations I made on the topic if desired.

The profiles I created were colorimetrically accurate to a fault. We used them for product photography under controlled lighting. The problem was that mathematical accuracy did not translate into pleasing visuals. Discontinuities in the profile translated into ugly transitions in images. Franz Herbert at Integrated Color Solutions took a different approach of prioritizing aesthetics over mathematical accuracy. That ended up working far better in the real world.

Cameras just aren't designed for accurate color. Scene referred imaging with perfect L-I imaging response still produces relatively unattractive images. However, for reproduction work this,  otherwise unattractive, is highly desired since the goal is to create a side-by-side matching copy.

However, even if a camera's imaging met L-I and "perfect (low dE)" prints could be made there are many reasons those may not match the original.

They could have different surface characteristics such as glossy or semi gloss characteristics that differ from the original. Colorimetric processes only retain accuracy when a print is illuminated at 45 degrees and there is no light from other angles that could reflect off a surface. A matte print w/o OBAs, displayed w/o glass, is impacted the least by this.

But even then a match might not occur unless the copy and original are displayed using D50 or a close spectral equivalent. This is because the reflected colors, simplified to the three CIE variables XYZ only match because the spectral characteristics of the original is not going to be the same. The two may be metameric at D50 (and 45 degrees, no uV, etc) but with a different illuminant metameric error will accrete.

And even with all that the CIE matching functions aren't perfect. Individual humans vary too. Sometimes quite a bit. Mostly this affects how two different people will see an image on two monitors, otherwise profiled identically, but with different backlight spectrums. Esp. CFL v LED. Print metameric error tends to be less as reflective spectra is usually much smoother than the monitor's RGB. It's especially noticeable on the shorter wavelength side from 440 to 500 nm where the "B" can match for one person but be off for another.

Here's a deep dive from a student at RIT about 20 years ago.

http://www.rit-mcsl.org/StudentResearch/mshaw/CMF_Thesis.pdf

Colorimetry works in a well controlled environment but there's a multitude of factors that can confound.
Logged

digitaldog

  • Sr. Member
  • ****
  • Online Online
  • Posts: 12505
    • http://digitaldog.net/
Re: New profile making software: Lumariver Profile Designer
« Reply #237 on: October 26, 2017, 02:57:57 PM »

Cameras just aren't designed for accurate color. Scene referred imaging with perfect L-I imaging response still produces relatively unattractive images. However, for reproduction work this,  otherwise unattractive, is highly desired since the goal is to create a side-by-side matching copy.
Wouldn't that result in two unattractive matching items?  ;D
Logged
Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/

Doug Gray

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 848
Re: New profile making software: Lumariver Profile Designer
« Reply #238 on: October 26, 2017, 04:41:27 PM »

Wouldn't that result in two unattractive matching items?  ;D

Yep, it just turns the first one ugly.  ;D

It's an interesting psychological effect. If you print a typical landscape picture processed normally, then hang it on the wall and take a picture of that, print and hang it next to the first, you'll like the first picture but the second generally looks pretty garish. Kodak, and others, have trained us:

Paul Simon: "KodaChrome, give us those nice, bright colors ...."
Logged

markanini

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 57
Re: New profile making software: Lumariver Profile Designer
« Reply #239 on: October 26, 2017, 05:40:54 PM »

I disagree, for camera profiles my experience has shown pleasant colors and accuracy can go together when sacrificing total accuracy, by targeting a range of important subject colors like caucasian skin tones(max 1 DE in this case). Color error models likely do a good job of predicting percetibilty of color shifts, but dont account for color shifts in skin tones looking more objectionable for human subjects when treated equally to non-skin tones.
Logged
Pages: 1 ... 10 11 [12] 13   Go Up