问题
This question addresses in particular the question of curve fitting in the context of color mixing of paints, pigments, etc.
I am trying to guess the required proportions of two paints, let's say "Brown" (B) and "white" (W) to get to a given lightness value L.
I have made a "calibration curve" in the same fashion as one does so for applying the Beer-lambert law in chemistry. However, the curve is not linear so I cannot use the Beer-Lambert law.
Here's what I've done :
(1)
I have measured the spectrum of paint samples for these proportions of mixture, labeled a, b, c, d, ... etc.
a >>> W = 1, B = 0 (pure white)
b >>> W = 63/64, B = 1/64
c >>> W = 31/32, B = 1/32
d >>> W = 15/16, B = 1/16
e >>> W = 7/8, B = 1/8
f >>> W = 3/4, B = 1/4
g >>> W = 1/2, B = 1/2
h >>> W = 0, B = 1 (pure brown)
And these are the spectral reflectance curves that I got :
If I pick-up one reflectance value at a given wavelength, e.g. 500 nm, I get this nice curve, where the x axis represents the proportion of white paint in the mix, and the y axis the reflected light at 500 nm :
I'd like to guess by interpolation how much white I need to arrive at a certain amount of reflected light.
(2)
I have tried to fit an exponential curve to the data with scipy.optimize.curve_fit but the fit is pretty poor:
What kind of function would fit the data closely?
回答1:
I'll expand my comment since nobody has answered.
From what I see in the figure, there is a pattern. The best way would be to fit a curve that fits that pattern as a whole. You can do this without any math using Eureqa (the free trial should be enough): http://www.nutonian.com/products/eureqa/
If you want to remain in python and fit an exponential distribution, you can do the following: How to do exponential and logarithmic curve fitting in Python? I found only polynomial fitting
So imagine you have for the wavelength 500nm the following values:
y = [10,20,30,30,50,60,70,80,90,100]
x = [0.,0.3,0.5,0.6,0.72,0.77,0.84,0.9,0.95,1]
Then the code to fit the exponential curve would be:
import numpy as np
import matplotlib.pyplot as plt
from scipy.optimize import curve_fit
def func(x, a, b, c):
return a * np.exp(-b * x) + c
popt, pcov = curve_fit(func, x, y)
In this case we get that a,b, and c are:
popt = array([ 7.1907744 , -2.62804994, 2.45029842])
So to get the value of reflected light at a certain x (for instance 0.2), you can do:
func(0.2, 7.1907744 , -2.62804994, 2.45029842)
Which is 14.61
But you say it's a bad fit, if you don't need a model, you can do the following: If you don't really care about having a model you can use this: https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.interp1d.html#scipy.interpolate.interp1d
from scipy import interpolate
f = interpolate.interp1d(x, y, kind="quadratic") #you can try different kinds of interpolation
And then to find a value (for instance x=0.2):
ynew = f(0.2)
Which is 6.549
Or to have many values so you can plot them: ynew = f(np.linspace(0,1,1000)
来源：https://stackoverflow.com/questions/41802871/good-interpolation-method-for-color-mixing