How to use a color bar to measure corresponding RGB values with Numpy or opencv

Question

I am trying to measure corresponding "real" values of an RGB image from an RGB color bar.

The color bar is several pixel large and I am wondering what is the best way to represent it. The solution I have currently is to average them along the width.

import cv2
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns

img = cv2.cvtColor(cv2.imread('data/dark_img.tif'), cv2.COLOR_BGR2RGB)
color_bar = cv2.cvtColor(cv2.imread('data/c_scale.tif'), cv2.COLOR_BGR2RGB)


# average colors over the bar width to get unique rgb values over the scale length
color_profile = color_bar.mean(axis=1, dtype=int)
scale_max = 1000  # color bar matches values between 0 and 1000
real_values = np.linspace(scale_max, 0, color_profile.shape[0])


def closest_rgb(img_rgb, color_profile_rgb):
    """get indices of color bar array closest to RGB values in imag"""
    img_rgb = img_rgb[:, :, np.newaxis, :]
    rgb_distances = np.sqrt(np.sum((img_rgb - color_profile_rgb) ** 2, axis=-1))
    min_indices = rgb_distances.argmin(-1)
    return min_indices


indices = closest_rgb(img, color_profile)
mapped_values = real_values[indices]

sns.stripplot(data=mapped_values)

Is there a more accurate way?

Answer 1

This may be what you are looking for

I guess no need to convert to HSV

import cv2
import numpy as np

box = cv2.imread(r'C:\somepath\blue_square.png')
strip = cv2.imread(r'C:\somepath\color_strip.png')

# Convert color to gray -> every color has its unique gray representation 
# y = kr*R + kg*G + kb*B where kr,kg,kb are unique
box_gray = cv2.cvtColor(box,cv2.COLOR_BGR2GRAY)
strip_gray = cv2.cvtColor(strip,cv2.COLOR_BGR2GRAY)

# take average of the box and set +- 10% tolerance to search
box_gray_avg = np.average(box_gray)
lo_lim = box_gray_avg*0.9
hi_lim = box_gray_avg*1.1

# lets match to the middle 1 pixel thick of the strip
rows, cols = strip_gray.shape
strip_thin_gray = strip_gray[:,cols//2]

# find the indexes where the match 
indx = np.where((strip_thin_gray>lo_lim) & (strip_thin_gray<hi_lim))

print(indx)

Output (array([ 98, 99, 100, 101, 102, 103], dtype=int64),)