I'm currently digging into color calculations/ conversions and also work with DeltaE.

Although the mathematical way might be correct, how can I get a better "human accepted" difference?

Sounds weird, but check out these three examples:

Test one:
Color one:
L 92.05
A -7.57
B 65.76

Color two:
L 85.66
A -6.07
B 65.95

Test 1

deltaE: 6.56

Test two:
Color one:
L 92.05
A -7.57
B 65.76

Color two:
L 81.81
A -21.06
B 65.86

Test 2

deltaE: 16.93

Test three:
Color one:
L 92.05
A -7.57
B 65.76

Color two:
L 87.07
A -2.1
B 94.58

Test 3

deltaE: 29.75

Color one is always the same for comparison. To put it simple: Test 1 is yellow, test 2 is green and test 3 is a darker yellow.

Is there a way to prefer hue over all other values? If I show these three colors a layperson, this person wouldn't say that green is more similar to the other yellow, just because it's a completely different hue.

I know, mathematically it's correct, and yes I also see why "the green" is more similar (I don't know the correct terms to describe it in english). But my endgoal is that a user can input a color and from a limited color palette the three best matching colors will be shown. As I said, green is not a good match for yellow in simple terms.

For testing I used my own deltaE calculator (that's currently not public) and to verify it this one

If you know a better place to ask this, please let me know.

Quick edit:
I know that three formulars for deltaE exists, I tested all three and the 1994 gives me the best result for my original idea and puts green at third place. The calculator I linked uses 1976 and my own calculator originally used 2000.

1 Answer 1


DeltaE values lose meaning as they get bigger. So while there is a huge difference between a delta E of 0.1 and 2 there is not much difference between deltaE of 2-10 other than clearly different colors. DeltaE values of 16 and 29 are just if you cant see that something is wrong with your eyes.

Simply your values are way too big to be useful.

This is simply because color is not a vector value. I mean sure we can represent it as a vector but the space makes no sense. So distances don't mean much. We can off course tune spaces for different things but at the end of the day if the change gets too big it loses meaning as a measure of distance because we don't know what exactly is the shortest path. Its a bit hard to figure out what it should be like if all you can do is compare 2 nearby values and you have a nonuniform space.

Also before you pass judgement on the values make sure your monitor is calibrated and values are within gamut. Monitors lie awfully lot. I was surprised when i fist calibrated my view of how much things changed form what they ought to be.

Theres some very current research that tries to rectify this but we are not there yet.

  • Thank you very much! I thought the same, the values are really big. But unfortunately our color palette is pretty small. Pantone to our 200 colors :D A lot gets lost on the way
    – jona
    Feb 4, 2022 at 15:50
  • @jona only delra e values that really make sense then is values less than 1
    – joojaa
    Feb 4, 2022 at 16:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.