Possible wrong calculation

[VinceKumar]

The delta_e_cie2000 function is giving me the wrong calculation of numbers. I was just checking with a calculator online to verify reproducibility. It might be that the online calculator is wrong, but it seems these colors are more similar than opposites.

pt = [7,6,4]
rgb = [35,35,35]

color1_rgb = sRGBColor(*pt)
color2_rgb = sRGBColor(*rgb)
color1_lab = convert_color(color1_rgb,LabColor)
color2_lab = convert_color(color2_rgb, LabColor)
delta_e = delta_e_cie2000(color1_lab, color2_lab)
delta_e

delta_e in this case is 87.9 online calculator says delta_e is 7.4

Happens in both 2.1.1 and 2.2

[blue-j]

RGB values have to go through XYZ before converting to Lab - not sure if colormath performs that for you or not? (I am new to the library and just assessing it and haven't even installed it yet!). Checking against a number of calculators, including Bruce Lindbloom's and EasyRGB, I get a delta E of 7.43 for your values. If you could try converting to XYZ first and then Lab and then calculating delta E 2000 that would be a great check. Also, by default the target illuminant will follow through (D65 in the case of sRGB) and maybe we should convert to D50 when going to XYZ first?

[KelSolaar]

it seems these colors are more similar than opposites.

Well they are actually very different, one is close to black:

That being said my own computations yield 7.4342281211262584

[KelSolaar]

@VinceKumar: Out of curiosity have you tried to divide your input colours by 255?

[blue-j]

Delta E 2k is very powerful and accurate model of our visual system and so a dark neutral color vs. nearly black should have a small delta. These are not very different colors perceptually. Around 7.4 looks correct to me for these values (and now we have three independent calculations amounting to 7.4).

[KelSolaar]

Delta E 2k is very powerful and accurate model of our visual system and so a dark neutral color vs. nearly black should have a small delta. These are not very different colors perceptually.

There are better colourspaces to perform colour differences, e.g. CAM02-UCS. In any case the colours are perceptually much different as per my above screenshot. It is quite undeniable :)

FYI, a CIE 2000 Delta E of 7.4 is a very high value, i.e. Much perception of colour difference : [1]

References

1. Yang, Y., Ming, J., & Yu, N. (2012). Color image quality assessment based on CIEDE2000. Advances in Multimedia, 2012. doi:10.1155/2012/273723

[blue-j]

We are splitting hairs here; I agree with you. dE of 7.4 is certainly noticeable.

One question, CIECAM02 is indeed the best, but how does one perform a delta E calculation with values in that space? I'm not aware of any method! Thanks in advance.

[KelSolaar]

We are splitting hairs here;

Well, I'm really just trying to be precise :)

One question, CIECAM02 is indeed the best, but how does one perform a delta E calculation with values in that space?

RGB --> XYZ ---> CIECAM02 (JMh) ---> CAM02-UCS (J'a'b') ---> Delta E Luo et al. (2006)

[blue-j]

Forgive me then for a collaborative parry. Delta E can be considered in two ways - if one is expecting/desiring a match, then your Yang/Ming/Yu table is accurate. But if one is comparing colors more broadly and not expecting a match, then 7.4 actually represents a small difference in the global space of possible differences.

Thank you so much for the Luo citation and pipeline! I'm eager to review this!

[blue-j]

I'll leave a link to the Luo et al. (2006) here for other's benefit in case anyone interested reads this thread: https://s3-us-west-2.amazonaws.com/4843ec7c-89cf-4d26-a36a-0e40ebc9a3a7/luo2006.pdf

[blue-j]

I believe I have figured out the problem @VinceKumar It is quite similar to @KelSolaar 's suggestion. When specifying the RGB values, you have to specify explicitly that they are 0-255 values like this:

sRGBColor(35, 35, 35, is_upscaled=True)

Without "is_upscaled=True" you get the wrong result.

[KelSolaar]

Side note: this parameter should really be something like is_8_bit.

[blue-j]

in case it is helpful:

from colormath.color_conversions import convert_color from colormath.color_objects import LabColor, sRGBColor from colormath.color_diff import delta_e_cie2000

rgb1 = sRGBColor(7, 6, 4, is_upscaled=True) lab1 = convert_color(rgb1, LabColor, target_illuminant='D50') rgb2 = sRGBColor(35, 35, 35, is_upscaled=True) lab2 = convert_color(rgb2, LabColor, target_illuminant='D50')
delta_e = delta_e_cie2000(lab1, lab2)

print (delta_e)

Output is 7.430843882682743

[blue-j]

Output is closer to @KelSolaar's above if no _targetilluminant transform is specified and sRGB's D65 is held throughout:

7.434515642877179

[KelSolaar]

python-colormath output is correct assuming one accounts for colour scaling as per https://github.com/gtaylor/python-colormath/issues/81#issuecomment-348909218