Perceptual Rendering Intent -- ProPhoto to sRGB

[jc1]

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.

I make use of the latest version of PatchTool to compute those numbers.  I have gone through the same exercise repeatedly and no discrepancy was found. My knowledge is that much.
 
As for Lab=(88 -79 81)  for ProPhoto RGB=(0 255 0), the Lab values were read out directly from the CS info screen.

Regards,
jc

[joofa]

I make use of the latest version of PatchTool to compute those numbers.  I have gone through the same exercise repeatedly and no discrepancy was found. My knowledge is that much.
 
As for Lab=(88 -79 81)  for ProPhoto RGB=(0 255 0), the Lab values were read out directly from the CS info screen.

Regards,
jc

No, that is not the issue. Think about distances here, after all that is what the "perceptual equality" notion is about, which you mentioned.

Sincerely,

Joofa

[jc1]

It sounds like your LUT for perceptual mapping were fixed, and you are adjusting the gamut primaries of the initial, intermediate host space.

This is an old V2 engine with no PRMG intelligence. As for the intermediate host space, I shall compare it with BetaRGB in subsequent stage.
 

"Referring to Photogamut RGB
You are needing two other color spaces, not just one"
Got your point.
 
 
"Channel clipping / saturation clipping (in the histogram) resulting from gamut conversion is unfortunately completely meaningless."
View is respected.


"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."
As said, perceptual rendering is not always the best solution, so does RelCol or other rendering algorithm.


Regards,
jc

[jc1]

No, that is not the issue. Think about distances here, after all that is what the "perceptual equality" notion is about, which you mentioned.

Sincerely,

Joofa

Subject might be too academic and I shall leave it alone for now.

On a side note, I am more interested in determining optimal primaries as much time was spent with trial and error. For instance, if 2 primaries are fixed, how to determine 3rd primary optimally, or fixed 1 and find the other 2.

Regards,
jc

[jc1]

Why jc1RGB is chosen as Intermediate Color Space?

Intermediate Color Space Consideration: BetaRGB vs jc1RGB

I shall try to present it as brief as possible. Believe me, it is tedious and uninteresting exercise.

1) Compute the RGB-Lab Profile data for ProPhoto, call it the Source.
    Convert the Souce to Destination space for both BetaRGB and jc1RGB
    I shall not go into details on how they were created.

2) How to interpret my data
    Refer to below attached tables, Table1 and Table2.

    The data colored in green is the RGB-Lab data for ProPhoto (source space)
    The data colored in red in Table1 is the RGB-Lab data for  BetaRGB (destination space)
    The data colored in red in Table2 is the RGB-Lab data for  jc1RGB   (destination space)

    The data colored in red is the converted data with AbsCol from ProPhoto space.

     They are of the following format:

For example: Red 1

ID Name    R          G       B         L*       a*       b*                  R        G        B        L*       a*       b*
1   Red1 255.000 0.000 0.000 60.608 139.169 104.497  --->  255.000 0.000 21.000 61.947 107.273 100.699  (BetaRGB) Table 1
                                                                             
                                                                          --->  255.000 52.000 0.000 58.764 124.016 98.898    ( jc1RGB)  Table 2

Above example shows Photo RGB=(255 0 0) is mapped to BetaRGB RGB = (255 0 21) with AbsCol
                                                     and is mapped to  jc1RGB  RGB = (255 52 0) with AbsCol

3) Examining the data sets
    a) When converting from ProPhoto to BetaRGB
        Red1 to Red6      RGB = ( 255 0 21)  these colors are indistinguishable after conversion
        Red7 to Red21    RGB = ( 255 0 22)  these colors are indistinguishable after conversion

       note: Red21 in ProPhoto is RGB = (235 0 0)
       
    b) When converting from ProPhoto to j1RGB
        Red4 and Red5  =  (255 49 0)          these colors are indistinguishable after conversion

       note: Red4 in ProPhoto is RGB = (252 0 0)
       As I mentioned earlier, further optimazation is still possible by moving primaries G & B, for perfectness.

4) Above was done for RED channel only.
    As for the GREEN and BLUE, the results are comparable for both BetaRGB and jc1RGB and I shall not repeat here.

5) Examine the data in the tables to convince yourself.




Regards,
jc

[joofa]

Subject might be too academic and I shall leave it alone for now.

That is okay. We can leave it for now. I must also say that I like the graphs and the analysis you do. It is so refereshing than endless messages on ETTR, which camera has how much SNR, etc. Your messages are welcoming relief at least for me.

However, the subject is not all academic IMHO. Think about some of the gamut maps that many programs show, the gamut volume calculations, the presentation of chromacity diagrams in books, and very importantly, is there really a need to go to a nonlinear space (say Lab) to claim the supposed "perceptual uniformity", while not properly analyzing linear spaces (say xyz, rgb, etc.)?

On a side note, I am more interested in determining optimal primaries as much time was spent with trial and error. For instance, if 2 primaries are fixed, how to determine 3rd primary optimally, or fixed 1 and find the other 2.

That is an interesting proposition and one can try to devise a methodology for that.

Sincerely,

Joofa

[jc1]

I must also say that I like the graphs and the analysis you do. It is so refereshing than endless messages on ETTR, which camera has how much SNR, etc. Your messages are welcoming relief at least for me.

Thank you for your words of encouragement.

Regards,
jc

[jc1]

Why jc1RGB is chosen as Intermediate Color Space? Part 2

Intermediate Color Space Consideration: BestRGB vs jc1RGB vs BetaRGB
 
 
BestRGB is another candidate, its Gamut volume if viewed with CTP3 is marginal larger than that for jc1RGB.


 
Performance Comparison
 
As there is no benchmark to follow, I use my method of comparison as outlined in previous post.
ProPhoto (Source) --> BestRGB (Destination)
                         --> jc1RGB   (Destination)
                         --> BetaRGB (Destination)
Rating
Rating 1 is better than Rating 2, and that Rating 2 is better than Rating 3.

The performance is summarized as follows.
 
      BestRGB   jc1RGB   BetaRGB        Remark
R       2              1           3            G=B=0
G       1              1           1            R=B=0
B       3              1           2            R=G=0
 

Regards,
jc

../ update graph

[Peter_DL]

On a side note, I am more interested in determining optimal primaries as much time was spent with trial and error.
…
ProPhoto (Source) --> BestRGB (Destination)
                         --> jc1RGB   (Destination)
                         --> BetaRGB (Destination)
Rating 1 is better than Rating 2, and that Rating 2 is better than Rating 3.

I think for me, and maybe for others as well, it is not so clear at which stage of progress with "trial and error" suggested procedure already is. Understandable though due to the complex nature of the subject.

As said, perceptual rendering is not always the best solution, so does RelCol or other rendering algorithm.

Another conclusion from my earlier pRGB-to-sRGB perceptual gamut compression exercises was that an relevant advantage with real-world images can only be produced by subsequent local contrast enhancement on the most saturated colors (versus straight RelCol + same post-processing). Isn’t it the core of this subject, to keep perceivable, graduated differences with highly saturated colors, smooth transitions + texture.

Regards, Peter

--

[jc1]

it is not so clear at which stage of progress with "trial and error" suggested procedure already is. Understandable though due to the complex nature of the subject.

The current beta 2.1 are working profiles.  I have stopped further development work and concentrating on analyzing the profiles' quality . If there is no negative feedback for the next 2 to 3 months, I may remove the beta status, but not too sure though at this point in time.

Regards,
jc

[jc1]

Intermediate Color Space Consideration
 
Color Saturation Comparison: BetaRGB vs jc1RGB
 
The saturation performance was examined with Gamutvision at L=0.5 (default setting) when converting from ProPhoto (Input or Source profile) to BetaRGB and jc1RGB (Output or Destination profiles)
 
If you have Gamutvision, you can do the same with the following settings.
1 ProPhoto       3 ProPhoto
2 BetaRGB        4 0_jc1RGB (beta2.1)
Rendering: Absolute, for both
Select: xy Chromaticity (Saturation map)
 
The following diagrams show how the  gamut boundaries (in 2D) are mapped between the source profile and  the detination profiles at L=0.5, where BetaRGB is on the left and jc1RGB is on the right.
 
Hints:
1) The gamut boundary for ProPhoto is represented with dotted lines (at different saturation levels)
    and that for the destination profiles (BetaRGB and jc1RGB) by solid lines.
 
2) The saturation analysis tellies with my previous tabulated result and my observation.
 
3) This implies "larger is better."
    http://www.brucelindbloom.com/BetaRGB.html
 
For more detailed explanation, kindly refer to http://www.gamutvision.com/docs/gamutvision_displays.html#CIE_xy
 

 

OT:
CP800 user, you might be interested in some of my work done which can be found here.

Regards
jc
 
 

[Bart_van_der_Wolf]

3) This implies "larger is better."
    http://www.brucelindbloom.com/BetaRGB.html

Hi JC,

You are mis-representing what Bruce Lindbloom says:

One important characteristic would be that the working space is suffiently large that it can properly encode (or contain) all colors that are important to an application. This implies "larger is better."

Another attribute, which conflicts with the above, is that the working space should be as small as possible, so that quantization errors may be minimized. This implies "smaller is better."

Bruce describes two conflicting attributes of an ideal workingspace, you pick one. By doing so, you allow larger quantization error for the 99% (?) most common colors, in order to accomodate a few outliers. What is your rationale for that choice? Do you encounter many real life (!)colors that are outside the BetaRGB gamut?

Cheers,
Bart

[jc1]

Hi Bart,

You are mis-representing what Bruce Lindbloom says:
Bruce describes two conflicting attributes of an ideal working space, you pick one.

I stated conclusively that " larger is better" and that is implying jc1RGB is the better choice as an intermediate color space.
Gamut for jc1RGB is about 16.3% (using gamut calculation with PatchTool and with simple arithmetic) larger than that for BetaRGB, hence "larger is better". I have no intention in proposing another working space.

My definition of an ideal intermediate space would be one that color difference and clipping is minimized when converting from one color space to the other or vice versa. Gamut volume is not the deciding factor.

By doing so, you allow larger quantization error for the 99% (?) most common colors, in order to accommodate a few outliers.

(I may not fully understood your concern but for purpose of discussion...let me try... )
If you are implying that if a larger working space is chosen than that for BetaRGB, then (severe or may be not that severe) quantization error may be introduced.  I think that may not be the main concern. My reasons are as follows.

Choice of gamma may has a direct influence in quantization error as gamma affects distribution of RGB points within the gamut. As both spaces use gamma 2.2,  and that the difference in gamut volume is not that great, I can't feel the effect. Frankly, I do not know how to compute it.  

Sometimes ago, I analyzed the error introduced  due to choice of 8-bit (per channel) against 16-bit (per channel), or vice verse, and with GRAXol2006_Coated_1 as Reference test data, I have the following result to share.


(note: unit in dE2000)

a) For sRGB, it is not critical if either 8-bit or 16-bit is used. Hence there is no good reason to use 16-bit for sRGB image.
b) For 16-bit, dE for Adobe RGB and Prophoto is comparable.

So, if 16-bit is used, dE is about the same for work space that is larger than Adobe RGB, be it BetaRGB, jc1RGB,  BestRGB or ProPhoto. I am not hinting that 16-bit must be used for aRGB image, the choice is greatly depends upon the user. I am unsure if above illustration is appropriate. Please note that above results are merely based on my test condition.

What is your rationale for that choice?

Which is more appropriate as an intermediate color space, BetaRGB or jc1RGB?

Do you encounter many real life (!)colors that are outside the BetaRGB gamut?

No. But if you are converting from ProPhoto to BetaRGB, it is possible that highly saturated Red may be clipped. I have demonstrated that with  ProPhoto, red channel will be clipped if  R>235 (B=G=0) and if BetaRGB is chosen as the intermediate color space, and that is what I am trying to avoid.  If this is not the concern, then use direct RelCol sRGB.

Have a nice day.

jc

.../spelling check

[jc1]

Matrix vs Cluts (Part 1)
 
My recommended approach for perceptual color space conversion from ProPhoto to sRGB is known to be "complex" . In actual fact, it is not, just that additional steps are involved.
 
Using my method, the image with ProPhoto profile is converted to jc1RGB, subsequently to sRGB_jc1 and finally to standard sRGB. The initial step is similar to converting to sRGB with straight RelCol. The main difference is that by converting to an intermediate color space first, the " color distortion" that could have been introduced due to conversion is significantly smaller, compare with directly convert to sRGB. This loss is further reduced to its minimal with jc1RGB as the choice of intermediate color space.
 
The matrix conversion has its advantages, not only that its profile size is relatively smaller and portable, color transformation can be carried out within the known xyz color space. The shortfall is, either sRGB, Adobe RGB, ProPhoto or other proposed color spaces, which are known to be Absolute color space, are not associated with mapping functions.
 
In V2 (icc profile specification version 2), color rendering with perceptual rendering intent in CMM can be carried with output type profile such as printer profile. Output profile uses Cluts ( Color Look Up Tables). ProPhoto has 2 of its primaries extended outside the visible locus and hence problem could arise if the image with ProPhoto has colors that are not within the visible locus. Are these invisible colors exist? I do not know. The fact is, if they are non existence, why we care?
 
I am going to show the clipping effect of converting ProPhoto to sRGB  with straight RelCol or matrix transformation, based on the following illustration. Your may have to refer back to my earlier thread, for how my tabulated data is interpreted.
 


Note: * these are clipped if import into PS.
 
Again, above only shows the RED channel. In general, there are less clipping effect on the GREEN, but could have more clipping effect on the BLUE if the blue components are outside the visible locus (non conclusive!). For BLUE, I noticed that these clipped  components that are falling outside the horse-shoe-shape could have their L (L in L*a*b*) values virtually zero.
 
Regards,
jc
 

[jc1]

Matrix vs Cluts (Part 2)
 
Color Degradation Analysis: RelCol vs Perceptual
 
Test Sample: 4096 test point as before.  <Reference Link>
 
Conversion Spaces
a) Source Space: jc1RGB
b) Destination Space: sRGB
 
Methodology
A)
1. Define reference
    Start with Source space, generate RGB's Lab values, our Reference data.
2. Convert from Source space to sRGB with straight Relcol. Compute Lab and compare it with Reference.
    Refer to first Diagram for result.
3. Convert from Source space to sRGB perceptually with sRGB_D50_jc1. Compute Lab and compare it with Reference.
    Refer to second Diagram for result.
 
B) Generate Tiff (RGB) Reference chart, import to PS. Assign profile with jc1RGB, that is our source image.
1. Convert from Source space to sRGB with straight Relcol. Import result to CTP3
2. Convert from Sources space to sRGB perceptually, use jc1 approach. Import result to CTP3.
    Results are shown with last Diagram.


 

 

 
 
Observations:
1) Refer to first 2 diagrams. No color degrading was observed.
2) Refer to last or third diagram.
    a) For conversion with RelCol, relative position of test points is well maintained. This is in comparison with test points in source space ( not shown).
    b) For conversion with Perceptual rendering, converted test points were re-distributed locally. More data points can be clearly seen with compare to that using convert with straight RelCol.
 
Comments
1) Above results show no color degrading with Perceptual rendering using output profile, sRGB_D50_jc1. Is conversion with perceptual rendering yields better color reproduction (with sRGB_D50_jc1) than with straight Relcol?
The observation is inconclusive, reason being that the sample size might be too small and that the RGB pitches of the test patches are too far apart. Further evaluation should be carried out with larger sample size or with real image. With real image, at your own time own target, anyone?
2) One thing for sure, more color details can be recovered with perceptual rendering, which has been clearly illustrated.
 

Regards,
jc

[jc1]

Inviting Comments on
 
If "complexity" is not the deciding factor, can jc1's approach replace straight Relcol, when converting from ProPhoto to sRGB?


Thank you.

Sincerely,
jc
 

.../rephrase question
 

[joofa]

Inviting Comments on
 
If "complexity" is not the deciding factor, can jc1's approach replace straight Relcol, when converting from ProPhoto to sRGB?


Thank you.

Sincerely,
jc
 

.../rephrase question
 

Unfortunately, JC1, it seems like that complexity is the issue here. There are some capable people here that contribute to the color forum. Unfortunately, it seems Iliah Borg is no longer contributing, but Peter_DL, DigitalDog, Schewe, and others are still here. Complexity of operations is a thing that people dread. But, it seems like you might have to distill your approach a little to make it more presentable and understandable. I think you want to say something useful here but the message is not coming across the way you want it.

Sincerely,

Joofa

[jc1]

Matrix vs Cluts (Part 3)
 
Space Conversion Analysis
 
Aim: To examine color reproduction capability with jc1RGB and sRGB_D50_jc1, by comparing with conversions to and from other popular color spaces such as sRGB, aRGB (Adobe RGB) and  PhotoGamut RGB.
 
Test sample: 5852 test points, covering from RGB=(0 0 0) to (255,255,255) with pitch of 15.
Color Matching: CIEΔE 2000, under D50 illuminant, 2° observer.
 
Results are summarized as follows.



 
Comment:
Data speak for itself.


.../re-grouping for clarify.

[jc1]

Unfortunately, JC1, it seems like that complexity is the issue here. There are some capable people here that contribute to the color forum. Unfortunately, it seems Iliah Borg is no longer contributing, but Peter_DL, DigitalDog, Schewe, and others are still here. Complexity of operations is a thing that people dread. But, it seems like you might have to distill your approach a little to make it more presentable and understandable. I think you want to say something useful here but the message is not coming across the way you want it.

Got your point.

Thanks.
jc

[jc1]

Photoshop action: Pro2sRGB_perceptual.atn
1-click action is available now.
Download the PS action from the below link.
 
Procedure
1. Load your ProPhoto image into CS.
2. Load PS action: Pro2sRGB_perceptual.atn
3. Shift F9 to execute the perceptual conversion from ProPhoto to sRGB
 
 
Download Link
1) Pro2sRGB_perceptual.atn
2) jc1RGB beta 2.1
 
All are in WinZip
Action was tested with CS4 under Windows xp3.
 
Do feel free to request if further details are needed.

Thank again Joofa for your valuable advice.
 
Regards,
jc
 

.../make sure the 2 profiles are installed correctly.