∆E 2000 vs ∆E CIELAB 1976 (skin tones)

[Magnus]

Hi, for years we have been using ∆E CIELAB 1976 calculation when measure color deviation on print and proofs (and displays sometimes). A year ago I was on a seminar with a color specialist from GMG who talked about ∆E 2000 as a successor of the old 1976 formula. He said that one of the advantage of ∆E 2000 is that it's not linear, but instead calculate the deviation of color more like the perception of the human eye (under the correct light conditions). I though this sounded like a good step in the right direction regarding color communication. Now ∆E 2000 has also been added in the new ISO 12647-7:2016 for proofing.

The last days I've been measuring prints and proofs with both ∆E CIELAB 1976 and ∆E 2000 to see if theory matches reality. Saturated full tones and gray balance are measuring as I expected. But I've noticed that skin tones is actually affected the opposite way than I expected. The ∆E-value of sensitive skin tones is actually lower when measuring with ∆E 2000. IMHO deviation in the hue of a skin tone is one of the main things that customers would take notice of.

When I look at the visual representation of ∆E 2000 vs ∆E CIELAB 1976 I looks like ∆E 2000 should be more sensitive than ∆E CIELAB 1976 in skin tones, but my tests give me the opposite results.

I've done all measuring with a brand new Techkon SpectroDens Premium New Generation (D50 2°) and also cross checked with an Xrite i1Pro 2.

What's your thought about this? Am I missing something? Do you got any other input?

Here's a link to my test file if you want to give it a try: https://www.dropbox.com/s/tvprgkd6czg70fe/Delta_E_Test.pdf?dl=0

 

[arossetti]

It is my understanding that DeltaE00 penalizes difference in lightness/hue more than chroma. Which mimics how the human visual system perceives color.

 

[Magnus]

It is my understanding that DeltaE00 penalizes difference in lightness/hue more than chroma. Which mimics how the human visual system perceives color.

Yes, exactly. Wouldn't that result in a higher ∆E value in skin tones thats deviates in hue compared to ∆E76?

 

[arossetti]

Yes, exactly. Wouldn't that result in a higher ∆E value in skin tones thats deviates in hue compared to ∆E76?

Can you provide the reference LAB values and the measurement?

 

[arossetti]

But also, I think this is also why the tolerances for DeltaE00 are tighter on most standards than DeltaE76.

In 12647-7 Primaries are DeltaE of 5
In 12647-7:2016 Primaries are DeltaE00 of 3.5

Right? I'm not sure if this variation in value is due to the math behind the formula for each DeltaE method or if the values are just not intended to be 1:1. I'm not the color scientist just the adopter

 

[Magnus]

Can you provide the reference LAB values and the measurement?

You can find the DeltaE values in the attached photo in my first post. I did not record the actual Lab-values.

 

[Stephen Marsh]

Here is an experiment… JPEG attached:

Stephen Marsh

 

[Magnus]

Here is an experiment… JPEG attached:



Stephen Marsh

Interesting Stephen, can you do the same experiment on the other patches? There is a URL to the PDF in my original post.

 

[Stephen Marsh]

Interesting Stephen, can you do the same experiment on the other patches? There is a URL to the PDF in my original post.

Sure Magnus, PDF attached.

PSO Coated v3 measured using Absolute Colorimetric intent.

View attachment DELTA.pdf


Stephen Marsh

 

[Magnus]

Sure Magnus, PDF attached.

PSO Coated v3 measured using Absolute Colorimetric intent.




Stephen Marsh

Thanks Stephen, your results is quite similar to mine (just a -+0,5 diff).

 

[arossetti]

To follow up on my previous point: "there is no mathematic equivalency between the ∆E*ab and ∆E00 metrics according to ISO 12647-2"

Tolerance Equivalency between ∆E*ab and ∆E00 Metrics By Lufei Yu. I don't have a copy of this dissertation but it might be an interesting read on the topic.

 

[Magnus]

I got this input from Juergen at GMG (I hope he doesn't mind me sharing) -

"Several researches have shown that in fact, the DE2000 formula is the closest to our perception. With 12647-7:2016 tolerances for the verification have been lowered in order to get an average comparable pass/fail, if you verify the many proofs. This means that from e.g. 1000 proofs where 50 failed with the old DE76 criteria, approx 50 will now fail with the DE2000 criteria. Most probably there will always be some proofs passing the one set but failing the other. So: the intention was not, to get sharper with the rule sets, but better (according to human perception). But: it is known that the DE2000 formula is not the 100% solution for all color areas. Researches have shown that from a visual aspect, some mid-saturated reds as well as some grey area are judged too lose so there are these areas where one would like to see more strict criteria."