Calman's dE numbers and Chromapure's dE numbers (Excel calc. sheet) do not match. - AVS Forum
Forum Jump: 
 
Thread Tools
post #1 of 21 Old 03-09-2013, 10:04 AM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
I really hoped this would not happen but I'm seeing some strange things:

First of all, you can find the sheet, under the heading ΔE, LSH, and RGB Analysis, here: http://www.chromapure.com/colorscience-calculators.asp

As you can see in screenshot from Calman, the measured x,y coordinates for white are x = 0.3188 y = 0.3248
The dE calculation then yields these numbers:
dE CIE 94 = 3.8659
dE CIEDE2000 = 6.0487

But, if you will them in the Excel sheet from Chromapure I'm getting CIE94 = 5.4 and CIEDE2000 6.9.

They can't both be right and I feel it is quite a large difference. How can this be?

Calman dE primary colors.PNG 43k .PNG file
Attached Images
File Type: png Calman dE primary colors.PNG (43.4 KB, 31 views)

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
Sponsored Links
Advertisement
 
post #2 of 21 Old 03-09-2013, 10:52 AM
AVS Special Member
 
Rolls-Royce's Avatar
 
Join Date: Apr 2003
Location: Victorville, CA
Posts: 1,981
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 6 Post(s)
Liked: 61
Quote:
Originally Posted by Jeroen1000 View Post

I really hoped this would not happen but I'm seeing some strange things:

First of all, you can find the sheet, under the heading ΔE, LSH, and RGB Analysis, here: http://www.chromapure.com/colorscience-calculators.asp

As you can see in screenshot from Calman, the measured x,y coordinates for white are x = 0.3188 y = 0.3248
The dE calculation then yields these numbers:
dE CIE 94 = 3.8659
dE CIEDE2000 = 6.0487

But, if you will them in the Excel sheet from Chromapure I'm getting CIE94 = 5.4 and CIEDE2000 6.9.

They can't both be right and I feel it is quite a large difference. How can this be?

Calman dE primary colors.PNG 43k .PNG file

CalMAN is likely using more decimal places for its calculations than it displays onscreen, while you are inputting the numbers CalMAN displayed into the spreadsheet. And I believe it may be using raw XYZ data in its calculations. If so, you have to take this rounding up or down as well the conversion from XYZ to x,y into account.

...Royce...

"I never drink...wine."
Bela Lugosi, DRACULA, 1931
Rolls-Royce is offline  
post #3 of 21 Old 03-09-2013, 11:24 AM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
Difference is too large for it to be a rounding error. Chromapure also rounds so you do need to take that into account when filling in the Excel sheet. But when I don't, it only makes for 0.1 dE difference.

Don't know about the raw XYZ conversion to x,y thing though. We are talking about +- 1.5 dE for the CIE94 formula. I'm not even comparing CP to Calman. I was (and am) just curious whether the dE calculation between both packages match.

On a side note, putting the numbers supplied by CP in the sheet, yields the same dE as reported in CP. Rounding can make a 0.1 difference as I said above.

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
post #4 of 21 Old 03-09-2013, 01:31 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,581
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 6 Post(s)
Liked: 163
You didn't take a 100% white reading in CalMAN, without it all our calculations are based off a 100cdm target.

Also by default in calman we normalize the target gamut luminance targets to the 75% white reading (still needs the 100% white to nonormalize). so that may also be a contributing factor.

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #5 of 21 Old 03-09-2013, 02:08 PM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
Correct I did not and it is admittetly a silly oversight. So I need to take that last white reading at 100% stimulus. I don't entirely understand the contributing factor you talk about.
Does it refer to taking absolute white reading at 100% and then it gets normalized but I do not quite understand what that entails?

I thought only x,y had to be measured for white and that Y has no bearing on x,y and is therefore not included for calculating its dE. In Tom's sheet it makes no difference for the dE number for white no matter what I fill in for Y. I can even leave it empty and the dE for white does not change as it only takes x,y into account.
Not trying to argue of course, I'll repeat my test and also measure 100% white.

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
post #6 of 21 Old 03-09-2013, 06:08 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,581
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 6 Post(s)
Liked: 163
Sound like chromapures calculator is incomplete then

Think for a second, 5% grey and white look very different and have the same x,y

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #7 of 21 Old 03-10-2013, 04:21 AM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
Hi Joel,

You said
Quote:
You didn't take a 100% white reading in CalMAN, without it all our calculations are based off a 100cdm target.

I did take a 75% stimulus reading for white.
What is Calman expecting for the CMS Calibration with the quick analysis? 75% stimulus 100% saturation patterns? Well, that is what I selected using following reasoning : when I measure 75% white, are not the Y targets for the primaries and secondaries calculated relative to the Y of 75% white I have just measured? What does the 100% reading change then? Just when I thought I had most of the theory down. Sorry for the probably obvious questions.

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
post #8 of 21 Old 03-10-2013, 06:03 AM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
Just because you all got me hooked on calibration, I've made my very own Excel sheet with raw data:-).
The thing I find peculiar is why the CIE94 formula in Calman does not produce the exact same results for dE than the CIE94 formula in the Excel sheet (which logically agrees with the Chromapure software as both are made by Tom Huffman).

None of what we do here is life threatening, but I did expect them to be much closer. Any and all opinions are valued:D





Differences with CIE94 formula.xlsx 75k .xlsx file
Attached Files
File Type: xlsx Differences with CIE94 formula.xlsx (75.1 KB, 12 views)

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
post #9 of 21 Old 03-10-2013, 08:09 AM
AVS Addicted Member
 
mark haflich's Avatar
 
Join Date: Dec 2000
Location: brookeville, maryland, usa
Posts: 19,185
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 53 Post(s)
Liked: 293
The autocal built into the Samsung SP-800 and SP900 projectors (Jor Kanes specifications) didn't require a reading for Y for white.

In theory the three Ys for the respective primaries should sum to 1 or 100% if correct, should they not.. Calibrated white Y is a sum of the three Y for the calibrated white and that sum obviously can be calculated rather than measured.

Of course to do calibrate the three Ys to the correct values (or as close as one is able) which are a set percentage a 100% stimulus white baseline would need to be measured initially to establish the 100% sum number. I hope I have this right if not I am sure the multitudes will rejoice in correcting me.

Mark Haflich
markhaflich@yahoo.com
call me at: 240 876 2536
mark haflich is online now  
post #10 of 21 Old 03-10-2013, 04:10 PM
AVS Addicted Member
 
mark haflich's Avatar
 
Join Date: Dec 2000
Location: brookeville, maryland, usa
Posts: 19,185
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 53 Post(s)
Liked: 293
I would add that with a front project you will run out of one primary first and the other two will be set to less than their max in order to allow the run out of primary to meet its correct percentage of the total for calibrated white. That's why a calibrated whitre will be less than the maximum uncalibrated white where the other two primaries are running higher than they should. Torch mode.

Mark Haflich
markhaflich@yahoo.com
call me at: 240 876 2536
mark haflich is online now  
post #11 of 21 Old 03-10-2013, 04:52 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,581
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 6 Post(s)
Liked: 163
The CMS's on projectors are build around the fact that light is additive.

If you have the x,y for white, and the x,y for red, green and blue at the same stimulus you can derive the normalized luminance levels for red green and blue. But that isn't particularly relevant to dE formulas.


If you look at any deltaE formula they are all based on converting your colormetric data into the L*a*b* color space to calculate from. In order to make that conversion you need to have the relative luminance of the value you are converting to L*a*b* and the reference white level in order to properly calculate the samples L*a*b* values. If you aren't including the luminance data, you aren't properly seeding all the information into the algorithms.


That is also why CalMAN needs both the 75% reading and the 100% reading. The 75% establishes the target values, but you still need the 100% value to properly normalize all the values going into the L*a*b* conversions.

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #12 of 21 Old 03-11-2013, 12:35 AM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,581
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 6 Post(s)
Liked: 163
I was away all weekend checking this on my phone.

After looking at the chromapure.xls worksheet, the discrepancy with their calculator is that it's only designed to work at 100% luminance.

If you run CalMAN with 100% luminance gamut targets, the numbers should agree.

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #13 of 21 Old 03-11-2013, 05:24 AM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
No worries, I consider it a privilege that you help people out on the forum.
I _could_ be wrong as I have to check first (I'm using both Chromapure and have just recently ventured into Calman), but the CMS work in CP is done at 75% stim. 100% sat.
Inputting those numbers (so without a 100% luminance reading for white) in the Excel sheet does give me dE's accurate to about 0.1.

But I would have to check that no 100% white reading was taken along the road.

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
post #14 of 21 Old 03-12-2013, 11:35 AM
AVS Special Member
 
TomHuffman's Avatar
 
Join Date: Jan 2003
Location: Springfield, MO
Posts: 6,353
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 14 Post(s)
Liked: 201
Whenever you have any question about whether a dE calculation for CIE94 or CIEDE2000 is correct, simply consult.

http://www.brucelindbloom.com/index.html?ColorCalculator.html

and

http://www.brucelindbloom.com/index.html?ColorDifferenceCalc.html

On the first page, set reference white to D65, type the xyY values and then click the xyY button.
This will provide conversions for all of the color formats. Focus on the Lab value. The Lab value for x0.3188, y0.3248, Y1.0 is 100, 5.3877, -0.5167

Next, copy, paste the Lab values from that page into the Lab Sample on the second page. In the Lab reference simply enter the Lab equivalent of whatever reference color you wish. For white (0.3127, 0.329, 1.0), that is 100, -0.0025, -0.0139.

The CIE94 dE for x0.3188, y0.3248, Y1.0 is 5.4. CIEDE2000 is 6.9.

This is exactly what ChromaPure reports. The discrepancy is based, I am guessing, on how CalMan treats luminance in dE calculations with white.

The industry standard is, and has always been, to NOT consider luminance for dE calculations when measuring white. All you look for is errors in chromaticity. The luminance data point is accommodated by gamma. This is why you always use unity as the Y value for white dE calculations. CalMan has an option for including luminance data in dE calculations for white, which yields unexpected results and will conflict with every other white dE calculation in this business with which I am aware. With color there should be no differences.
prsut likes this.

Tom Huffman
ChromaPure Software/AccuPel Video Signal Generators
ISF/THX Calibrations
Springfield, MO

TomHuffman is offline  
post #15 of 21 Old 03-12-2013, 01:43 PM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
This debate seems to have its roots way back already:-) http://www.avsforum.com/t/1087084/de1976-or-de1994-for-calibration/30
Admittedly, that debate is well above my level of comprehension on the matter, however in my defense, I was blisfully unaware this is so convoluted. I can certainly understand it can become somewhat annoying whenever a "newbie" comes to town asking the same thing you have answered so many a times.

I did finally make sense of the lightness vs luminance thing but more reading is required before I can join talks about Lab Luv and consorts.

But, what is sure is that Calman uses Lightness (or should say luminance) data for calculating the dE of D65 white and CP does not. So I agree this is most likely the source of the discrepancy.

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
post #16 of 21 Old 03-30-2013, 01:49 PM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
Thank you for those links Tom. Much appreciated. Sorry it took me this much time to come back to this.

Could someone check these numbers I've tested please?

For Red
x = 0.6480
y = 0.3301
Y = 15.0681

White Y (75% stim) was measured at 66,4397 (cd/m² I believe).

Putting these numbers in Bruce Lindboom's site yieds a CIE94 dE of 2.247185
Calman reports a CIE94 dE of 1,9070

In order to input all the numbers I required, I had to calculate 1 number myself. The Y for red is 21.26% of the Y of white if you're talking reference numbers. For the above numbers I found the Y of Red is 22.67% of that of white. I hope I did that correctly:).

So, my question for Calman is, why are you calculating things differenly and what is the reasoning behind this?

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
post #17 of 21 Old 03-30-2013, 02:11 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,581
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 6 Post(s)
Liked: 163
Quote:
Originally Posted by Jeroen1000 View Post

Thank you for those links Tom. Much appreciated. Sorry it took me this much time to come back to this.

Could someone check these numbers I've tested please?

For Red
x = 0.6480
y = 0.3301
Y = 15.0681

White Y (75% stim) was measured at 66,4397 (cd/m² I believe).

Putting these numbers in Bruce Lindboom's site yieds a CIE94 dE of 2.247185
Calman reports a CIE94 dE of 1,9070

In order to input all the numbers I required, I had to calculate 1 number myself. The Y for red is 21.26% of the Y of white if you're talking reference numbers. For the above numbers I found the Y of Red is 22.67% of that of white. I hope I did that correctly:).

So, my question for Calman is, why are you calculating things differenly and what is the reasoning behind this?

What is the 100% white reading?

In order to convert XYZ to L*a*b* you need to normalize the XYZ values against the Y value for white. Chromapure omits this and is wrong for it.

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #18 of 21 Old 03-30-2013, 02:23 PM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
The white reading was

x= 0,3171
y= 0,3247
Y= 124,9846

So, the calculator on Bruce's side is also not entirely accurate?

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
post #19 of 21 Old 03-30-2013, 03:20 PM
AVS Special Member
 
sotti's Avatar
 
Join Date: Aug 2004
Location: Seattle, WA
Posts: 6,581
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 6 Post(s)
Liked: 163
Quote:
Originally Posted by Jeroen1000 View Post

The white reading was

x= 0,3171
y= 0,3247
Y= 124,9846

So, the calculator on Bruce's side is also not entirely accurate?
Bruces Calculator is completely accurate.

If white Y is 124.98 and your white reading is 15.06 cd/m then that reading has a Yn value of .1204 Yn

If you are in CalMAN and using 75% white as your target luminance (the default setting), then your target Y for red is the 75% white reading * 0.2126 or in this case 14.125

When you normalize the target to white you get 0.113 as the Yn target.

so now we have

Measured x,y,Yn value of 0.648, 0.3301, 0.1204
Target x,y,Yn value of 0.64, 0.33, 0.113

So using D65 as our white point we get

Measured L*a*b* 41.27, 67.52, 59.74
Target L*a*b* 40.08, 64.87, 54.43

The dE94 between those two L*a*b* values is 1.9


I'm not entirely sure what Tom does that raises his dE numbers, but when you correctly normalize luminance you'll get the correct result.
Jeroen1000 likes this.

Joel Barsotti
SpectraCal
CalMAN Lead Developer
sotti is online now  
post #20 of 21 Old 03-31-2013, 03:12 AM - Thread Starter
Senior Member
 
Jeroen1000's Avatar
 
Join Date: Nov 2008
Posts: 314
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
Thank you for your excellent reply as usual. I feel this is important to know for people using both calibration packages. The discrepancy always appears to be in the normalizing luminance part. I'm not keen on starting a debate on who is correct but I am surprised that there different views on applying (as it appears to me) a fixed formula.

Surely something to take into account.

My gear: Panasonic TH-42PF11EK pro plasma display. -- Iscan Duo video processor -- i1 display 3 colorimeter -- i1 pro 2 spectrometer
Jeroen1000 is offline  
post #21 of 21 Old 04-01-2013, 09:45 PM
AVS Special Member
 
TomHuffman's Avatar
 
Join Date: Jan 2003
Location: Springfield, MO
Posts: 6,353
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 14 Post(s)
Liked: 201
As I have said previously, I think it is a mistake to use this "normalizing" method for the grayscale, so nothing more needs to be said about that.

However, with color it is a more complicated question. Technically speaking, neither of us do 75% color precisely the correct way (for 100% stimulus color, both are the same). This would require having the user select a preferred gamma and then using that gamma value to establish the correct target based on a 100% white reference (For example, 0.113 would not be the correct luminance target in your example if you had selected a gamma of 2.4). This is the method we use for establishing targets for the Lumagen 125-point color cube calibration, and I am sure they do the same for their LUT calibrations.

Having said that, I honestly think that both methods are acceptable for basic 7-point gamut calibration at 75%. The CalMan method is closer to the theoretically correct approach and our method is the more commonly used approach by video engineers, or at least so I am told (I asked one). The reason I say that both are acceptable is that the difference between a CIE94 value of 1.9 and 2.2 is negligible. However, if you were calibrating using 50% stimulus color (which very few do) then the difference would be bigger, and our dE value would be considerably larger than the CalMan value. However, even here the only consequence for our users would be that they would be prompted to calibrate to tighter tolerances than he or she otherwise would.

So, to be clear, the small differences you have seen are not related to the dE calculation itself, but rather to the method used for establishing the correct target against which the sampled color is compared.

Tom Huffman
ChromaPure Software/AccuPel Video Signal Generators
ISF/THX Calibrations
Springfield, MO

TomHuffman is offline  
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