Pages: 1 ... 6 7 [8]   Go Down

Author Topic: Phase One Trichromatic In Depth Article with Raw Files  (Read 16466 times)

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 688
    • Hikes -more than strolls- with my dog
Re: Phase One Trichromatic vs Standard Back Matrices
« Reply #140 on: January 13, 2018, 03:28:37 AM »

So, to eliminate as many variables as possible, it is probably best practice to fix the whitepoint, and – to avoid as many cans of worms as possible – I personally prefer to use illuminant E. The latter simplifies spectral data conversions for the colorpatches. No need to know the spectral data of the original illuminant, or obfuscate the results with daylight spectra.

That's interesting Oscar.  So you compute the matrix that best converts white balanced raw data to reference xyz/Lab values resulting from the target being lit by illuminant E?  Then perform a chromatic adaptation to D65 for a/sRGB?

Since the matrix varies with (the inverse of) color temperature, how accurate does it come out compared to trying to estimate it for the correct illuminant as I did?  Maybe I should try it out.

Pardon all the questions, I am self-taught and there are very few people whom I can talk to about this (:-)

Jack
Logged

opgr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1921
Re: Phase One Trichromatic vs Standard Back Matrices
« Reply #141 on: January 13, 2018, 04:35:33 AM »

So you compute the matrix that best converts white balanced raw data to reference xyz/Lab values resulting from the target being lit by illuminant E? 

Yes. I specifically converted the Babel color spectral data to XYZ using NO illuminant spectrum. That results in the cleanest XYZ-E reference for my purposes. I can make those XYZ colors available if you'd need them.


Then perform a chromatic adaptation to D65 for a/sRGB?

Yes, although I convert to ICC D50 using straight multiplication for adaptation. Then let colorsync figure out conversion to whatever the user selects as outputspace. I may change this later to incorporate colorclipping control. The trick then is to adapt the outputmatrix to illuminant E by simply dividing out its original illuminant.

I prefer to use simple division/multiplication for illuminant conversions because during several steps of the process it is already done that way. i.e. whitebalancing the raw data is done by multiplication, Lab to XYZ is generally defined with multiplication. Using bradford transforms
has a disputable theoretical advantage if you allow 2 degree observer logic to assess a 10 degree observer problem for extreme conversions. That advantage is however not of the precision that would radically change the desired results, especially in smaller conversions.

Since the matrix varies with (the inverse of) color temperature, how accurate does it come out compared to trying to estimate it for the correct illuminant as I did?  Maybe I should try it out.

Some thoughts:
the algorithm trying to find the matrix uses reference XYZ values. If those values are given with D50 or D65 illuminant, then the algorithm should end up finding D50 or D65 illuminant, since you already whitebalanced the data.

If you have the spectral data of the patches, and the spectral data of the original scene illuminant, then you could theoretically adapt the patches using spectral conversion, though I doubt that the entire process recognises that kind of precision, and the objective of colormatching is to generalise the matrixes so they can easily be adapted under normal PCS circumstances.

That is: we're trying to find matrixes that are optimised for use in XYZ and straight illuminant conversions.

Note also that CCT is not an equivalence. That is an unfortunate mistake perhaps introduced by RAW converters. CCT is a one-way conversion, and very sensitive, it may be insightful to try and figure out the size of the differences in XYZ vs the corresponding size of the differences in CCT (and then try to translate that to a perceptually uniform error space). But that would be (euphemistically) left as an exercise for the reader...
Logged
Regards,
Oscar

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 688
    • Hikes -more than strolls- with my dog
Re: Phase One Trichromatic vs Standard Back Matrices
« Reply #142 on: January 13, 2018, 08:34:13 AM »

Yes. I specifically converted the Babel color spectral data to XYZ using NO illuminant spectrum. That results in the cleanest XYZ-E reference for my purposes. I can make those XYZ colors available if you'd need them.

Ok, I have the spectral data in their spreadsheet. In fact I calculate the XYZ/Lab reference values by multiplying them by the SPD of the relative D illuminant, calculated from an estimate of the CCT.

Yes, although I convert to ICC D50 using straight multiplication for adaptation. Then let colorsync figure out conversion to whatever the user selects as outputspace. I may change this later to incorporate colorclipping control. The trick then is to adapt the outputmatrix to illuminant E by simply dividing out its original illuminant.

I prefer to use simple division/multiplication for illuminant conversions because during several steps of the process it is already done that way. i.e. whitebalancing the raw data is done by multiplication, Lab to XYZ is generally defined with multiplication.

Ok, so in the case of illuminant E the adaptation matrix is just the destination white point XYZ coordinates in a diagonal I assume.

Using bradford transforms has a disputable theoretical advantage if you allow 2 degree observer logic to assess a 10 degree observer problem for extreme conversions. That advantage is however not of the precision that would radically change the desired results, especially in smaller conversions.

Some thoughts:
the algorithm trying to find the matrix uses reference XYZ values. If those values are given with D50 or D65 illuminant, then the algorithm should end up finding D50 or D65 illuminant, since you already whitebalanced the data.

Yes, with the proviso that with my  procedure if the data is not perfect (and it never is) the best compromise matrix may shift the white point slightly (typically a couple/few tenths of deg. K).  But I find that acceptable given the fact that, unless we were able to obtain the SPD of the illuminant (and as photographers we normally don't) we typically do not know the original CCT to that level of precision.

If you have the spectral data of the patches, and the spectral data of the original scene illuminant, then you could theoretically adapt the patches using spectral conversion, though I doubt that the entire process recognises that kind of precision, and the objective of colormatching is to generalise the matrixes so they can easily be adapted under normal PCS circumstances.

That is: we're trying to find matrixes that are optimised for use in XYZ and straight illuminant conversions.

Right.

Note also that CCT is not an equivalence. That is an unfortunate mistake perhaps introduced by RAW converters. CCT is a one-way conversion, and very sensitive, it may be insightful to try and figure out the size of the differences in XYZ vs the corresponding size of the differences in CCT (and then try to translate that to a perceptually uniform error space). But that would be (euphemistically) left as an exercise for the reader...

Challenge accepted for the next rainy November day :)

Thanks a lot Oscar, you've been very helpful.
Jack
Logged

Jim Kasson

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1955
    • The Last Word
Re: Phase One Trichromatic vs Standard Back Matrices
« Reply #143 on: January 13, 2018, 10:58:35 AM »

That's interesting Oscar.  So you compute the matrix that best converts white balanced raw data to reference xyz/Lab values resulting from the target being lit by illuminant E?  Then perform a chromatic adaptation to D65 for a/sRGB?

That leaves you open to illuminant metameric failure. Adaptation != Illumination.

Jim

opgr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 1921
Re: Phase One Trichromatic vs Standard Back Matrices
« Reply #144 on: January 14, 2018, 08:25:09 AM »

That leaves you open to illuminant metameric failure. Adaptation != Illumination.

Jim

Yes, ceci n'est pas une pipe

However, we are not trying to find a single set of primaries for all types of illuminants. We are trying to find a single set of primaries for a single type of illuminant, and preferably adaptation within the range of that single type of illuminant is accomplished with simple multiplication and division in PCS.

From the user perspective, I would like to select the sensor response for the type of illuminant. Daylight vs Tungsten for example. What I certainly do NOT want is a RAW converter that decides to interpolate between Daylight and Tungsten primaries based on my abuse of a colortemperature parameter which is what I am offered for colorcast appearance. I'm fairly certain that errors introduced by that logic are far greater than the errors of adaptation by multiplication in PCS.






Logged
Regards,
Oscar

ErikKaffehr

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 10952
    • Echophoto
A lot of excellent info from Jack and Oscar...
« Reply #145 on: January 14, 2018, 10:19:54 AM »

Hi,

I would like to put forward a lot of gratitude to Jack (Hogan) and Oscar without forgetting Jim (Kasson). There is a very nice forum on DPR with folks much brighter than me:

https://www.dpreview.com/forums/1061

Some recent discussions there were based around SMI (Sensitivity Metamerism Index ) and it sort of correlated with the discussion about the IQ 3100 Thrichromatic.

https://www.dpreview.com/forums/post/60253992

It is quite interesting to follow Jack's reasoning.

I still have a presumption that some of the difference may stem from differences in IR and UV filtering strategies between the two backs. In my humble opinion, there are three interesting indications pointing in that direction.

  • The chlorophyll greens are greener in the images from the thrichromatic. That may be a result from stronger IR-filtration.
  • Doug's description of the Thricromatic did mention UV/IR filtration - especially in connection with lime greens.
  • Uzman Dawood's comparison of the IQ3100Mp with the Thrichromatic. Uzman's examples show axial chroma that is not present on the Thrichromatic. https://fstoppers.com/originals/phase-one-iq3-100mp-trichromatic-colors-compared-209065

The samples Dave has provided indicate a large difference in lime greens. Personally, I don't have access to modern Phase One backs, although I happen a P45+. Another interesting device I happen to have in my possession is a Sony Alpha 900 which is regarded to have the best compromise in CFA design by a few knowledgeable folks, like Iliah Borg and TheSuede. I also happen to have an A7rII.

I did make a small test with all three cameras and all three reproduced lime green reasonably well, at least with studio flash, within JDI I would say.

Could it be that the IQ3100MP has a bit 'soft' IR-filtering? It could be a compromise towards use with large beam angle lenses. The same could apply to UV, of course.

The interesting part here is that it seems that vegetable greens normally have very high IR-content, while the foliage patch on the ColorChecker has low IR. So a profile generated from the ColorChecker could perfectly well reproduce the vegetation patch on the CC while be quite a bit off on real world vegetation. Would that have been the case it would not be the first time. Both the Nikon D200 (I think) and the Leica M8 had weak IR filters, AFAIK.

I would guess on stronger IR or UV filtering on the Thrichromatic being an improvement.

The easiest way to find out would be to shoot with an IR cut off filter, but that would mean that new profiles would be needed.

I would be very happy making those tests if someone loaned me an IQ3100MP and a Thrichromatic in combination with a camera body and a lens.

Best regards
Erik
Logged
Erik Kaffehr
 

billthecat

  • Full Member
  • ***
  • Offline Offline
  • Posts: 104
Re: Phase One Trichromatic In Depth Article with Raw Files
« Reply #146 on: January 14, 2018, 11:50:53 AM »

I'd like to say that I have enjoyed this thread and have learned a lot. After many months it is still a mystery what the Trichromatic is. The layman's explanations have been more confusing to me than the scientific ones.

Looking at Usman Dawood color checker samples between the Canon and Phase his colors don't match on the color checkers. I used the 3DLut creator to match using his TIFFfiles and the color checkers matched well between the two cameras. I don't really know what I'm doing but I'm working on learning. When I matched the purple issue didn't seem to matter too much. I don't know if the marker is more green or yellow.

I would assume the color checkers should match first before you compare, and then if they don't could that be a reason for color issues?

This is from a PetaPixel article. https://petapixel.com/2018/01/12/get-medium-format-colors-full-frame/

I just noticed that the Fstoppers article was comparing the Phase to Phase, not a Canon.

I attached a JPEG of my results in Adobe RGB.
« Last Edit: January 14, 2018, 11:56:58 AM by billthecat »
Logged

andrewrodney

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 13229
    • http://www.digitaldog.net/
Re: Phase One Trichromatic In Depth Article with Raw Files
« Reply #147 on: January 14, 2018, 01:12:59 PM »

The layman's explanations have been more confusing to me than the scientific ones.
Stick with the scientific ones! Marketing speak is a much bigger, moving target.
Logged
Andrew Rodney
Author “Color Management for Photographers"

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 688
    • Hikes -more than strolls- with my dog
Re: Phase One Trichromatic In Depth Article with Raw Files
« Reply #148 on: January 15, 2018, 03:08:58 AM »

You may very well be right on all points Erik, I guess we'll have to wait for someone with a monochromator and the two backs to draw us their CFAs.

Jack
Logged

Jack Hogan

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 688
    • Hikes -more than strolls- with my dog
Re: Phase One Trichromatic vs Standard Back Matrices
« Reply #149 on: January 15, 2018, 03:27:40 AM »

From the user perspective, I would like to select the sensor response for the type of illuminant. Daylight vs Tungsten for example. What I certainly do NOT want is a RAW converter that decides to interpolate between Daylight and Tungsten primaries based on my abuse of a colortemperature parameter which is what I am offered for colorcast appearance. I'm fairly certain that errors introduced by that logic are far greater than the errors of adaptation by multiplication in PCS.

Good points Oscar, they will be interesting to investigate on that rainy November Saturday.

Jack
Logged
Pages: 1 ... 6 7 [8]   Go Up