Perceptual Rendering Intent -- ProPhoto to sRGB

[bjanes]

Please kindly explain the components of these profiles (matrix primaries, white point, TRC / gamma, ...)
and the idea behind the double conversion starting with AbsCol.

Thanks.
& Best regards, Peter
--

In his response JC showed some custom RGB matrix profiles. I can't follow the complexities of his workflow, but it should be noted that matrix profiles lack the lookup tables necessary for perceptual rendering. There are many problems with current perceptual rending algorithms since they apply arbitrary compression of the color gamut without actually looking at what color values are in the image. For a narrow range gamut, no compression at all may be needed. Ver 4 ICC profiles are supposed to handle perceptual rendering, but are not widely available. The purpose of this post is to invite further discussion.

Regards,

Bill

[digitaldog]

Imaginary values may be produced via mathematical transformation.
Adobe color model is based on CIELAB 50 hence imaginary color cannot  be created in that way.

And yet the chromaticity diagrams you yourself have provided here show two colors (primaries) that are imaginary.

[digitaldog]

Ver 4 ICC profiles are supposed to handle perceptual rendering, but are not widely available.

Right, and ideally, both profiles in the chain are V4 and implement the PRMG. I don’t know if this is the case here.

[Peter_DL]

  
  
JC , - the Coding Efficiency of ProPhoto RGB is 87.3%, the Coding Efficiency of the Lab gamut is just 35.1%. The Coding Efficiency % indicates the relative portion of the encoding space that represents real colors. Some of the larger volume working spaces contain many RGB (or Lab) triplets for which there is no physical counterpart, and therefore could be considered wasteful.
Source:  www.brucelindbloom.com

How comes? The percentage with Lab is quite poor. The Lab Gamut is essentially the same as CIEXYZ. Going back to the original color matching experiment done with CIE RGB primaries it was found that a negative r matching function is needed to match colors outside the limited CIE RGB triangle. To avoid such negative matching  function, a mathematical construct was derived, CIE XYZ which bears all positive matching functions, at the cost of primaries XYZ which are not real colors.
Reference: http://www.fho-emden.de/~hoffmann/ciexyz29082000.pdf

The ProPhoto RGB gamut which is a subset of CIEXYZ  is likewise bearing "imaginary" primary colors. These are needed to hold other saturated colors, particularly yellow hues which actually DO exist in reality.

Not sure if this is a most accurate or comprehensive way of description,
however, it may meet your question.
 

Ver 4 ICC profiles are supposed to handle perceptual rendering, but are not widely available.

Never got these profiles to work for me.

The purpose of this post is to invite further discussion.

+1, same with me.

Bill, actually I’m quite happy with the PhotoGamut RGB pathway (+ local contrast enhancement on the most saturated colors) as outlined in this earlier discussion also referenced above. Depending on image. Others are better served with straight RelCol, pRGB to sRGB.

Regards,

Peter

--

[TylerB]

I have found utilizing PhotoGamut at some point in a workflow to be helpful in certain circumstances, primarily in working down to a space smaller than itself, sRGB for an on demand book for example. Unfortunately some colors currently available in the gamut of the newest inkjets available are not within PhotoGamut, and therefore some available colors from the printer/paper/inkset my never be utilized. An update of PhotoGamut considering contemporary printer gamut might be welcome.
The discussion and experimentation is useful though. One of th ongoing problems, under discussed, is getting from large working spaces down to output or web spaces in a viable manner. There is still no really good way to do it, and often massaging colors into eventual gamut before conversion is a frustrating task given the usual tools- Hue/Sat, Selective Color, Replace Color, etc- often seem to create as many problems as they solve. Tools for putting as cow's heads on a pig continue to improve, but color editing seems a bit in a rut.  V4 and smart cmms seem promising but still a pipe dream, and potentially have new unforseen problems as well....
So, seems like a good thread.
Tyler

[jc1]

Hi,

AFIK, when you assign a profile in Photoshop (versus Convert to a profile) CIE LAB conversion is not used, rather the RGB values in the file are left unchanged and only the associated profile information (colour primaries and tone curve) is changed. Hence, as Peter stated, it is possible to produce imaginary colours in the image.

Here is a test I have done to confirm this (all files are 8-bit TIFF).

• Firstly I created an sRGB file and filled it with 0-R 255-G 0-B a real colour in sRGB (file: sRGB.tif).
• I then assigned the ProPhoto RGB profile to the file (file: ProPhoto_assigned.tif). When the colour is checked it is still 0-R 255-G 0-B which in ProPhoto RGB is entirely imaginary.
• Finally I took the original sRGB file and converted it to ProPhoto RGB (file: Converted_to_ProPhoto.tif). As expected this did not change the appearance of the colour on-screen but it did change the colour values to 138-R 237-G 78-B which in ProPhoto RGB is the same colour as 0-R 255-G 0-B in sRGB.

This confirms that Assigning can create imaginary colours.

Regards
Nigel

When an image is assigned or tagged with a profile, the corresponding lab value will have to be re-determined. Logically speaking, it must be done with an internal color model. I hope someone more knowledgeable in PS can verify that.

It is true that ProPhoto RGB=(0, 255 0) in xyz is imaginary.
But, you may be wrong by simply stating that the Lab values for RGB=(0 255 0) in PS is imaginary.


Where is ProPhoto RGB=(0, 255 0)?



a) Green_1:   XYZ=(13.519161 71.187450 0.000005) --> Lab=(87.58 -186.69 150.99)
b) Green_1a:  Lab=(87.58 -186.69 150.99) --> Lab=(87.58 -128 127)  (simply clipped it)
c) Green_1b:  Lab=(88 -79 81)

Green_1:   An imaginary color
Green-1a: To show direct clipping is different from colorimetric transformation.
Green_1b: The Lab value in PS is (88 -79 81) and it is a realizable color.


sRGB --> ProPhoto (smaller gamut --> Bigger gamut)
When converting sRGB to ProPhoto, Lab value remains unchanged. Since this point has the same Lab value, it must be of the same color.


jc

.../clarification 1
.../clarification 2

[jc1]

In his response JC showed some custom RGB matrix profiles. I can't follow the complexities of his workflow, but it should be noted that matrix profiles lack the lookup tables necessary for perceptual rendering. There are many problems with current perceptual rending algorithms since they apply arbitrary compression of the color gamut without actually looking at what color values are in the image. For a narrow range gamut, no compression at all may be needed. Ver 4 ICC profiles are supposed to handle perceptual rendering, but are not widely available. The purpose of this post is to invite further discussion.

Regards,

Bill

Hi Bill,

"JC showed some custom RGB matrix profiles"
a)  jc1RGB is matrix profile
b) sRGB_D50_jc1 is output type profile
The technical aspects were already given.
 
"can't follow the complexities of his workflow"
Sorry if it sounds complex but it is not my intention.
 
" it should be noted that matrix profiles lack the lookup tables necessary for perceptual rendering"
sRGB_D50_jc1 is an output profile with Cluts. Its colorimetric accuracy AbsCol /RelCol is on par with standard sRGB. It has been clearly shown.
 
"There are many problems with current perceptual rending algorithms since they apply arbitrary compression of the color gamut without actually looking at what color values are in the image"
Over here the scope is well defined, ProPhoto to sRGB.
 
"For a narrow range gamut, no compression at all may be needed. Ver 4 ICC profiles are supposed to handle perceptual rendering, but are not widely available"
Can't agree more. Unsure in practice if anyone uses V4 for perceptual compression?
 
"The purpose of this post is to invite further discussion"
Yes and please try it. It is a working model unless someone can prove it otherwise.
 

Best regards
jc

.../typo err

[jc1]

And yet the chromaticity diagrams you yourself have provided here show two colors (primaries) that are imaginary.

If that chromaticity diagram was incorrectly presented, I shall try to clarify.

Right, and ideally, both profiles in the chain are V4 and implement the PRMG. I don’t know if this is the case here.

PRMG model is a nice concept. But it may or may not work for matrix with imaginary primaries as in the case of ProPhoto.
I am using V2, it is simply based on practical applied engineering current available. It is nothing fanciful here.
 

Best regards
jc
 
 

[jc1]

Hi,

 
JC , - the Coding Efficiency of ProPhoto RGB is 87.3%, the Coding Efficiency of the Lab gamut is just 35.1%. The Coding Efficiency % indicates the relative portion of the encoding space that represents real colors. Some of the larger volume working spaces contain many RGB (or Lab) triplets for which there is no physical counterpart, and therefore could be considered wasteful.
Source:  www.brucelindbloom.com

Brucelindbloom's hard work is much appreciated.
Practical implementation for Lab encoding causes the wastefulness. The waste if can be recovered at all, it may not carry any significant impact on real color reproduction, but it may render color conversion less painful.

 
"How comes? The percentage with Lab is quite poor. The Lab Gamut is essentially the same as CIEXYZ. Going back to the original color matching experiment done with CIE RGB primaries it was found that a negative r matching function is needed to match colors outside the limited CIE RGB triangle. To avoid such negative matching  function, a mathematical construct was derived, CIE XYZ which bears all positive matching functions, at the cost of primaries XYZ which are not real colors.
Reference: http://www.fho-emden.de/~hoffmann/ciexyz29082000.pdf".

There is distinctive difference between Lab and XYZ in that Lab is perceptually equal distance, XYZ is not. Lab is derived from XYZ, no?

The ProPhoto RGB gamut which is a subset of CIEXYZ  is likewise bearing "imaginary" primary colors. These are needed to hold other saturated colors, particularly yellow hues which actually DO exist in reality."

Unless there is direct transformation (XYZ) from RAW, I can't figure out where else can the imaginary color comes from. The practical implementation in the real world has put multiple barriers to restrict it from happening, from tiff (V6 spec) to PhotoShop CS, for instance.

Thanks

jc

[digitaldog]

If that chromaticity diagram was incorrectly presented, I shall try to clarify.

Its not wrong. It shows that two primaries of ProPhoto RGB are imaginary colors, they fall outside the chromaticity diagram. Technically they are not colors (humans can’t see em).

PRMG model is a nice concept. But it may or may not work for matrix with imaginary primaries as in the case of ProPhoto.

The PRMG is what makes the V4 profiles kind of useful!

[joofa]

There is distinctive difference between Lab and XYZ in that Lab is perceptually equal distance, XYZ is not.

Unfortunately, an incorrect interpretation of the XYZ color space has also resulted in it being labeled as perceptually bad. There is a new realization now regarding what went wrong in the original interpretation.

Sincerely,

Joofa

[Nigel Johnson]

But, you may be wrong by simply stating that the Lab values for RGB=(0 255 0) in PS is imaginary.

Please do not misquote me. If you bother to read my post you will see that I stated that RGB=(0 255 0) is imaginary for ProPhoto RGB and is a real colour in sRGB and I do not state that an arbitrary RGB=(0 255 0) will be imaginary. The purpose of my test was to demonstrate that assigning a ProPhoto RGB profile to an sRGB image could create imaginary colours.

When a profile is assigned in Photoshop the RGB values in the file are not changed and NO internal colour model is used. Therefore if you create a ProPhoto RGB test image from an sRGB file by assigning the ProPhoto profile you may possibly create an image with imaginary colours (you could probably check if the image actually contains imaginary colours by using ColorThink or a similar utility). Hence, when you compare your complex process for converting from a large gamut (ProPhoto RGB) to a smaller gamut (sRGB) against a standard single profile conversion you may be showing differences that are purely a result of the imaginary colours in your test image.

sRGB --> ProPhoto (smaller gamut --> Bigger gamut)
When converting sRGB to ProPhoto, Lab value remains unchanged. Since this point has the same Lab value, it must be of the same color.

This only applies if you use convert to profile (my third bullet and example file) it does not apply if you use assign profile.

Nigel

(Edited so quotes show original source)

[Peter_DL]

... Therefore if you create a ProPhoto RGB test image from an sRGB file by assigning the ProPhoto profile you may possibly create an image with imaginary colours (you could probably check if the image actually contains imaginary colours by using ColorThink or a similar utility). Hence, when you compare your complex process for converting from a large gamut (ProPhoto RGB) to a smaller gamut (sRGB) against a standard single profile conversion you may be showing differences that are purely a result of the imaginary colours in your test image.

Nigel, - this is perfectly what I meant
(but did not reach to make clear). Many thanks.


JC, - if we can come over this,
I’d be interested what exactly determines the gamut primaries of your initial host space jc1RGB. If you need a somewhat smaller, handier space than ProPhoto RGB to tailor the subsequent perceptual conversion, why not using something known like e.g. Bruce Lindbloom’s Beta RGB. However, you mentioned that you can’t find an easy way out … (?).

By the way, it is summer (at least in the northern hemisphere). Many colorful flowers out there, to shoot and test…

Regards, Peter

--

[jc1]

Unfortunately, an incorrect interpretation of the XYZ color space has also resulted in it being labeled as perceptually bad. There is a new realization now regarding what went wrong in the original interpretation.

Sincerely,

Joofa

Hi,

Thank you for sharing. More pointers please.

Regards,
jc

[jc1]

Hi Nigel,

Thanks and welcome for sharing your thought with us.
Sorry if that was misunderstood. Let's move on.

I shall clarify if I did not make clear on thread: http://www.luminous-landscape.com/forum/index.php?PHPSESSID=24205b19480ced199ad3e9e3a10ee9ce&topic=56978.msg462413#msg462413

Regards,
jc


.../Just found
 I regret a mistake was made on a sentence in that thread, the below added words in red were what I intended to include originally.

But, you may be wrong by simply stating that the Lab values for RGB=(0 255 0) for ProPhoto in PS is imaginary.

[jc1]

I’d be interested what exactly determines the gamut primaries of your initial host space jc1RGB. If you need a somewhat smaller, handier space than ProPhoto RGB to tailor the subsequent perceptual conversion, why not using something known like e.g. Bruce Lindbloom’s Beta RGB. However, you mentioned that you can’t find an easy way out … (?).

By the way, it is summer (at least in the northern hemisphere). Many colorful flowers out there, to shoot and test…

Regards, Peter

BetaRGB was among my choices as an intermediate color space. I have had tested a few profilers too, including ArgyllCMS and i1Profiler with some simulated color test patches but result is unsatisfactory, and that certainly can be further improved. Each profiler has it strength and color rendering can be quite difference, where saturated colors are concerned. Most commercial profilers work great normally but the weakness is that there is no provision for fine tuning. Though i1Profiler has great performance, due to copyright, I cannot release the profiles to the general public.

If I will to post anything on that, I have to be better prepared on technical issue.

Great! The last time I had been to the west coast was 30 years ago on  an company assignment (non photography related and I didn't even has a camera with me). Beautiful and lovely place it was!

Now over here it is hot weather plus rainy season.

Regards,
jc

[jc1]

I have found utilizing PhotoGamut at some point in a workflow to be helpful in certain circumstances, primarily in working down to a space smaller than itself, sRGB for an on demand book for example. Unfortunately some colors currently available in the gamut of the newest inkjets available are not within PhotoGamut, and therefore some available colors from the printer/paper/inkset my never be utilized. An update of PhotoGamut considering contemporary printer gamut might be welcome.

With good printer profile, high color reproduction is still possible.
When image was converted from one color space to another color space with profiles and with perceptual intent, colors will be shifted due to cmm mapping process, that is inevitable.

If there are profiles to convert  perceptually to say sRGB (Beta2.1) or Adobe RGB, do we need another color space?

I am still trying hard to understand the audience here.

Thank you for feedback.

Regards,
jc

[jc1]

Hi all,

I was not presenting a new invention, but just trying to convince the audience that the conversion from ProPhoto to sRGB can be performed perceptually with the current cmm but with a different approach. The aim is to improve the color re-generation process.  Certainly, there may have other better way to do it that I am unaware of.

I do not have a real and good example to show. I had used the hypothetical model (see Example 2) which covers all the most saturated colors (be it R, G or B) in ProPhoto to demonstrate its capability. My supplemented photo images may not represent the best examples, but I still think they do not mislead the reproduction capability of this conversion process.

Let's move on by looking at it from a different perspective.

PhotoGamutRGB's Perceptual vs. jc1's Perceptual

I am using just one example here. If you have example to show that PhotoG does perform better, you are encouraged to show your image and with histograms if possible.



1. Original Tiff image
    Nikon raw --> Import into NX2 with ProPhoto profile --> Cropped and output to Tiff --> Import into CS

Following conversions were performed in PS.
2. a) Convert from Tiff to photoGamut
    b) Further convert to sRGB with RelCol sRGB

3) Convert with jc1's  2-step* suggestion.

Differences
For PhotoGamut, it was first converted perceptually to a much smaller space than that for ProPhoto, and then Relcol to sRGB.

As for my approach, the image was first AbsCol (or RelCol) to a intermediate color space which is smaller than ProPhoto but still significantly larger compared to sRGB. Then to sRGB with perceptual intent.


Notes:
a) * I shall keep using the same term to avoid further confusion.
b) Compare by observing the shadow detail as shown in the histograms.

Thank you.

jc

[joofa]

More pointers please.

Sign of the times - there was a time when my word would suffice. 

Take your example of color = [0,255,0], convert it to xyz and compare and see if you spot a discrepancy there. This is a little off topic to your OP so if you want then we can take it offline.

Sincerely,

Joofa

[Peter_DL]

BetaRGB was among my choices as an intermediate color space. I have had tested a few profilers too... Though i1Profiler has great performance, due to copyright, I cannot release the profiles to the general public.

I don't understand this reply.
It sounds like your LUT for perceptual mapping were fixed, and you are adjusting the gamut primaries of the initial, intermediate host space.
 
 
Referring to Photogamut RGB

If there are profiles to convert  perceptually to say sRGB (Beta2.1) or Adobe RGB, do we need another color space?

You are needing two other color spaces, not just one.
 
 

b) Compare by observing the shadow detail as shown in the histograms.

Channel clipping / saturation clipping (in the histogram) resulting from gamut conversion is unfortunately completely meaningless.

Just think about a chessboard pattern of gray vs a highly saturated color. Straight RelCol conversion from a large source to a small target space will most likely beat any perceptual gamut compression algorithm. Saturation clipping - in this general context - does not disclose the degree of image degradation and loss of perceivable texture.

Visually I find it hard to see the differences with the above illustration,
however, this may also have to do with the details of preparing these screenshots for posting.

Regards, Peter

--