Closed Santinell closed 9 years ago
Interesting. Care to send a pull request?
I don't know true reducing coefficient, but [L*0.5, a, b ] works better on my test colors. I will try to explore it more.
Reducing the "L" is not enough improves accuracy for my data :( In wiki for color difference recommends using CIEDE2000 instead of CIE76 (Euclidean distance) I found one implementation: https://github.com/markusn/color-diff
color-diff looks nice. Have you used it yet?
Yes, but factor of rising accuracy for both algorithms is: the larger colors in palette - the more accurate definition of the color. So now I choosing between using color difference and using some trees for storing colors (octree, kd-tree) and searching nearest neighbor.
IMHO for determine color in L_a_b color space - Lightness is least important So I think is needed reduce the L in Lab, for reduce its impact