Alternate Observers - AVS Forum
Forum Jump: 
 
Thread Tools
post #1 of 14 Old 12-05-2013, 08:42 AM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 4,558
Mentioned: 24 Post(s)
Tagged: 0 Thread(s)
Quoted: 141 Post(s)
Liked: 329
So I've been experimenting with the effects/differences between calibrating using the standard observer and some alternate observer color matching functions. I'd like to know if other calibrators use these functions at all and if so, what is your opinion on when and why one should use something other than the standard observer? For background on why and how some of these functions were developed I recommend reading Mark Shaw's Masters thesis. He also produces his own optimized set of color matching functions (Shaw-Fairchild).

I chose 3 different sets representing a range of differences as seen through the eyes of the standard observer with the results below. Each set of measurements was obtained from the display while a 3dLUT of the corresponding alternate observer was active and the probe was set to the 1931 2 degree standard observer. To create the LUTs themselves the probe was operated using the alternate color matching XYZ functions.

3dLUT #1: Judd & Vos



3dLUT #2: Shaw & Failrchild



3dLUT #3: 1964 10 degree (recommend by the CIE for observer subtense angles greater than 4 degrees)




The Judd & Vos based calibration appears cooler with more saturated blues and less saturated reds and yellows. Average color checker errors are 2.99 so the color space movement is fairly significant and one can visually observe the differences in actual material.

The Shaw & Fairchild based calibration is for all purposes identical to the standard observer. Average color checker errors are 0.72 (compared to 0.67 for the standard observer).

The 1964 10deg. based calibration has both a hue shift (yellow towards green) and some increased saturation of reds and blues and pull towards higher green saturation. Again there is a significant enough shift in color space (CC dE00 = 2.31) and grayscale to create easily identifiable changes in viewing material.

Personally I found the 1964 10deg. calibration shift toward green objectionable, especially for skin tones, however the Judd & Vos based calibration produced a subjective improvement in the material I have viewed so far. Perhaps this data set better matches my own color sensitivity?


EDIT:

As noted below by Sotti and Spacedriver, the experiment shown above is not correct as an example of "observer matching" because the underlying spectral distribution of the rec709 primaries is not part of the specification. One can only perform the matching at the white point where the spectral distribution is specified. In order to do this properly one must define "deviate" observer functions that attempt to align the underlying spectral distributions through least-squares (or other) statistical means.

-----

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.
zoyd is online now  
Sponsored Links
Advertisement
 
post #2 of 14 Old 12-05-2013, 09:10 AM
AVS Special Member
 
Chad B's Avatar
 
Join Date: Oct 2002
Location: Piqua, OH
Posts: 2,069
Mentioned: 3 Post(s)
Tagged: 0 Thread(s)
Quoted: 57 Post(s)
Liked: 349
Thanks for bringing this up and for the info. I have been experimenting with them as well. I'm on my phone now but when I get some time I will share my still limited impressions.

touring calibrator
Jeti 1211

Latest reviews:


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.
,
To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.
,
To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.

Chad B is online now  
post #3 of 14 Old 12-05-2013, 01:02 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,621
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 28 Post(s)
Liked: 167
I'm not sure you're doing this right.

When you change your standard observer you need to change your targets as well.

There are not currently well defined standards for alternate CMFs for rec709, since it primaries are simply given as x,y in 1931 2 degree.

We can calculate new x,y, locations for D65 based on alternate CMFs since D65 has a spectral definition.

Here are some of the coordinates for D65 with alternate CMFs:
D65 with StandardObserver_2_Degree
x 0.3127115954, y 0.3290084044

D65 with StandardObserver_10_Degree
x 0.3138053150, y 0.3309763442

D65 with JuddAndVos_ModifiedCIE_2_Degree
x 0.3159923048, y 0.3350360907

D65 with CIE_170
x 0.3135398703, y 0.3309292947

D65 with CIE_170_Schanda_Csuti
x 0.3143893616, y 0.3307062293

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #4 of 14 Old 12-05-2013, 02:16 PM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 4,558
Mentioned: 24 Post(s)
Tagged: 0 Thread(s)
Quoted: 141 Post(s)
Liked: 329
Quote:
Originally Posted by sotti View Post

I'm not sure you're doing this right.

When you change your standard observer you need to change your targets as well.

There are not currently well defined standards for alternate CMFs for rec709, since it primaries are simply given as x,y in 1931 2 degree.

We can calculate new x,y, locations for D65 based on alternate CMFs since D65 has a spectral definition.

Thanks sotti, are you assuming that I have measured everything using 1931_2 CMFs? In that case yes I would need to calculate new x,y targets for each observer and yes that would require knowing the spectral distribution produced by my display for each primary. But that's not what I have done, the measurements use the tabulated alternate CMFs convolved with the measured spectra as part of the probe driver to generate new XYZ values. So the probe takes care of all the heavy lifting. The targets remain the same and the 3dLUT is adjusted to match so that the modified x,y values hit rec709.

-----

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.
zoyd is online now  
post #5 of 14 Old 12-05-2013, 02:26 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,621
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 28 Post(s)
Liked: 167
No, you still aren't picking up what I'm putting down.

Lets say you are using an i1 Pro in CalMAN.

In CalMAN you can select your CMF, ie do the heavy lifting of modifying the XYZ data.

Now when you measure using an alternate CMF you get different data back, but what numbers do you calibrate to? .3127, .331 for D65, no you don't!

D65 if you are using Judd Voss is at 0.316, 0.335, so if you want to try and get your white point correct, those are the numbers you should target. But what about red? it's only defined as 0.64, 0.33, we don't have a spectral definition for red, so we can't go back to the spectrum and apply the Judd Voss CMF to that data. So there are no structured guidelines on what to do and we lack a concrete physical specification to fall back on.

If you just use an alternate CMF to calibrate a display to rec.709, it will look wrong. rec.709 has the 1931 2 degree observer baked into it's targets.

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #6 of 14 Old 12-05-2013, 02:52 PM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 4,558
Mentioned: 24 Post(s)
Tagged: 0 Thread(s)
Quoted: 141 Post(s)
Liked: 329
Quote:
Originally Posted by sotti View Post

Now when you measure using an alternate CMF you get different data back, but what numbers do you calibrate to? .3127, .331 for D65, no you don't! D65 if you are using Judd Voss is at 0.316, 0.335, so if you want to try and get your white point correct, those are the numbers you should target.

oh dear, it's sinking in now. So you are saying that a standard observer viewing a white patch at 0.3127, 0.3290 will perceive the same color as a JV observer viewing a white patch at 0.316, 0.335?
Quote:

But what about red? it's only defined as 0.64, 0.33, we don't have a spectral definition for red, so we can't go back to the spectrum and apply the Judd Voss CMF to that data. So there are no structured guidelines on what to do and we lack a concrete physical specification to fall back on.

If you just use an alternate CMF to calibrate a display to rec.709, it will look wrong. rec.709 has the 1931 2 degree observer baked into it's targets.

I'm still confused as to why it's just not a simple matrix transformation from one observer frame to another as long as your spectral distribution is the same for each measurement (i.e. you are not trying to color match an LCD in one observer frame to a plasma in another).

-----

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.
zoyd is online now  
post #7 of 14 Old 12-05-2013, 03:12 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,621
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 28 Post(s)
Liked: 167
Quote:
Originally Posted by zoyd View Post

oh dear, it's sinking in now. So you are saying that a standard observer viewing a white patch at 0.3127, 0.3290 will perceive the same color as a JV observer viewing a white patch at 0.316, 0.335?
That's exactly what I'm saying.

When you start looking at things like the ASTM documentation on SPD to XYZ, you'll see they label the 1931 2 degree as X, Y, Z and the 1964 10 degree as X10, Y10, ZX10 to differentiate the values. XYZ data generated with alternate CMFs are not equivalent.

So the spectral definition of D65 is a measure of a physical thing like length, if you have two things that are both 1ft long, no mater what unit of measurement (cm, inches, fathoms) you use, both things will measure the same. So it is with spectral data, if you have two things producing identical spectral signatures, any observer will report that they are identical. Since our perception of color is soft, the CMFs seek to correlate our perception of color to the physical spectrums. So much like changing units with length, the difference in measuring color with an alternate CMF is similar to the difference in inches or centimeters. You have to measure the original physical data and the physical data under test with the same units of measurement for the comparison to be valid, in this case the physical data is a spectral distribution.
Quote:
Originally Posted by zoyd View Post

I'm still confused as to why it's just not a simple matrix transformation from one observer frame to another as long as your spectral distribution is the same for each measurement (i.e. you are trying to color match an LED in one observer frame to a plasma in another).

That would be true, but what is the spectral data used to generate the rec.709 targets?
The most obvious answer would be a CRT spectrum, but nobody has sanctioned that as the canonical process.

What is being done in studios is the trusted device, usually a CRT, is measured in 1931 2 degree and calibrated to rec.709. Step 2 is to measure the original display with an alternate CMF. Then a secondary display, typically an OLED, is calibrated with the same alternate CMF to the x,y values that are created in step 2. You could then remeasure the OLED with 2 degree and create simple x,y offsets for the alternate CMF for that display.

But with a single display you could create four color matrix and transform a colorimeter from a 1931 2 degree device into a device measuring with an alternate CMF.
ConnecTEDDD likes this.

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #8 of 14 Old 12-05-2013, 03:34 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,621
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 28 Post(s)
Liked: 167
It should be noted, that the one color that is consistent between CMFs is illuminant E (flat spectral response).

By design any CMF given illuminant E should produce equal parts X and Y and Z, thus producing an x,y 0.333,0.333

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #9 of 14 Old 12-05-2013, 04:02 PM
AVS Special Member
 
derekjsmith's Avatar
 
Join Date: Oct 2003
Location: Mukilteo, WA
Posts: 1,890
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 1 Post(s)
Liked: 98
Here is the white paper we wrote on this last year: Issues in color matching
ConnecTEDDD likes this.

Derek

CTO / Founder - SpectraCal Inc.
derekjsmith is offline  
post #10 of 14 Old 12-05-2013, 04:43 PM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 4,558
Mentioned: 24 Post(s)
Tagged: 0 Thread(s)
Quoted: 141 Post(s)
Liked: 329
Thanks guys! I definitely don't want to reinvent the wheel here.
Quote:
What is being done in studios is the trusted device, usually a CRT, is measured in 1931 2 degree and calibrated to rec.709. Step 2 is to measure the original display with an alternate CMF. Then a secondary display, typically an OLED, is calibrated with the same alternate CMF to the x,y values that are created in step 2. You could then remeasure the OLED with 2 degree and create simple x,y offsets for the alternate CMF for that display.

So aside from the white point which can be calculated, the target primaries in an alternate observer frame need to be transferred via a CRT. So this process transfers what the studio regards as reference color reproduction to the new technology using CMFs better suited to do such a transfer given the spectral distribution of the new technology (OLED). This still isn't what I am after though. What if the reference CRT appears incorrect to a specific colorist and that this colorist has cone sensitivities better modeled by an alternate CMF, how does he calibrate his display using that alternate CMF so that he also sees correct rec709 colors?

-----

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.
zoyd is online now  
post #11 of 14 Old 12-05-2013, 04:57 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,621
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 28 Post(s)
Liked: 167
Quote:
Originally Posted by zoyd View Post

What if the reference CRT appears incorrect to a specific colorist and that this colorist has cone sensitivities better modeled by an alternate CMF, how does he calibrate his display using that alternate CMF so that he also sees correct rec709 colors?

Well these are Color Matching Functions.

These alternate CMFs help us make two different spectra appear the same, match, to more people in a more uniform way than the older ones did. They will never make it match for everyone.

If a certain person doesn't agree that a sony CRT calibrated to rec.709 with the 1931 2 degree observer is the reference standard, I don't know what to tell him. That was hardware in widest use when the standard was created. Something has to be the reference, if a single individual doesn't like the way they perceive the reference, that's like saying I think a kilogram is too heavy.

The only way to get around this is to move to a multi primary system where we more accurately record and reproduce the actual spectrum. Only when you actually replicate the physical thing will everyone agree that they match. Which is something is actively being researched by people like David Long http://www.cis.rit.edu/fairchild/PDFs/PRO39.pdf

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #12 of 14 Old 12-05-2013, 05:14 PM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 4,558
Mentioned: 24 Post(s)
Tagged: 0 Thread(s)
Quoted: 141 Post(s)
Liked: 329
Quote:
Originally Posted by sotti View Post

Well these are Color Matching Functions.

These alternate CMFs help us make two different spectra appear the same, match, to more people in a more uniform way than the older ones did. They will never make it match for everyone.

If a certain person doesn't agree that a sony CRT calibrated to rec.709 with the 1931 2 degree observer is the reference standard, I don't know what to tell him. That was hardware in widest use when the standard was created. Something has to be the reference, if a single individual doesn't like the way they perceive the reference, that's like saying I think a kilogram is too heavy.

smile.gif That was an extreme example of course, what I'm talking about is observer variability and the fact that people can fall into different groups where different CMFs perform better for these groups as described here. I asked Graeme Gill this question awhile ago and he seemed to think it was theoretically possible to calibrate in this fashion (measure using alternate CMFs and target rec709) but that the CMFs would need tweaking. Here is the quote:
Quote:
a CMF is just intended to allow computing metamers for that observer, not for cross matching between observers. You can transform a CMF by any 3x3 matrix and it will still perform perfectly as a CMF. But to transform between CMFs with minimal error, the CMFs need "aligning" in some fashion to best match their XYZ/Lab values, and since they will never align perfectly, it is a compromise. In the literature I've come across two approaches to align observer :- one is to minimize the least squares between the CMF curves themselves, and the other is to minimize the weighted delta E of a set of "typical" spectral test colors. Typically you will want to add a positivity constraint to the resulting CMFs too. You could make an argument for increased weighting of the Y/L matching.

-----

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.
zoyd is online now  
post #13 of 14 Old 12-05-2013, 09:12 PM
Advanced Member
 
spacediver's Avatar
 
Join Date: May 2013
Location: Toronto
Posts: 840
Mentioned: 2 Post(s)
Tagged: 0 Thread(s)
Quoted: 71 Post(s)
Liked: 70
Quote:
Originally Posted by zoyd View Post

So I've been experimenting with the effects/differences between calibrating using the standard observer and some alternate observer color matching functions. I'd like to know if other calibrators use these functions at all and if so, what is your opinion on when and why one should use something other than the standard observer?

I haven't read through this whole thread, but as I understand it, the standard observer function specifies a standard for metameric spectral sources, and thus allows a wide variety of primaries to be mixed in to achieve perceptually identical results.

Of course, there are issues with this - for one, the data was derived from a rather small number of subjects. Also troubling is that Grassman's additivity laws may not actually hold up all that well in many conditions.

But if it turned out that one could find a better standard observer than the one in use currently, one would need to redefine the CIE X Y Z colorspace (and the derivative spaces such as CIELAB), and the whole chain of video production, from the camera, to the colorist, to the calibrated end user display, would have to be based on this new colorspace.

And if indeed this standard observer function was better, the benefit would be that colors would appear more consistent across a variety of display primaries. It doesn't make sense to say one standard observer function appears bluer or greener, as far as I understand the issue.
spacediver is online now  
post #14 of 14 Old 12-06-2013, 05:56 AM - Thread Starter
AVS Special Member
 
zoyd's Avatar
 
Join Date: Sep 2006
Location: Planet Dog
Posts: 4,558
Mentioned: 24 Post(s)
Tagged: 0 Thread(s)
Quoted: 141 Post(s)
Liked: 329
Here is once such realignment of the Stiles-Burch CMFs based on the technique of Nayatani.

-----

To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.


To view links or images in signatures your post count must be 0 or greater. You currently have 0 posts.
zoyd is online now  
Reply Display Calibration

User Tag List

Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page


Forum Jump: 

Posting Rules  
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off