Change Color In Rgb Images

I have an image (I have it in both NumPy and PIL format), and I want to change where RGB values are [0.4, 0.4, 0.4] to [0.54, 0.27, 0.07]. By doing this, I want to change the road

Solution 1:

You can try this numpy approach:

img[np.where(img == (0.4,0.4,0.4))] = (0.54,0.27,0.27)

Solution 2:

I have to agree with Majid Shirazi regarding the proposed solution by Quang Hong. Let's have a look at:

idx = np.where(img == (0.4, 0.4, 0.4))

Then, idx is a 3-tuple containing each a ndarray for all x-coordinates, y-coordinates, and "channel"-coordinates. From NumPy's indexing, I can't see a possibility to properly access/manipulate the values in img in the desired way using the proposed command. I can rather reproduce the exact error stated in the comment.

To get a proper integer array indexing, the x- and y-coordinates need to be extracted. The following code will show that. Alternatively, boolean array indexing might also be used. I added that as an add-on:

import cv2
import numpy as np

# Some artificial image
img = np.swapaxes(np.tile(np.linspace(0, 1, 201), 603).reshape((201, 3, 201)), 1, 2)
cv2.imshow('before', img)

# Proposed solution
img1 = img.copy()
idx = np.where(img1 == (0.4, 0.4, 0.4))
try:
    img1[idx] = (0.54, 0.27, 0.27)
except ValueError as e:
    print(e)

# Corrected solution for proper integer array indexing: Extract x and y coordinates
img1[idx[0], idx[1]] = (0.54, 0.27, 0.27)
cv2.imshow('after: corrected solution', img1)

# Alternative solution using boolean array indexing
img2 = img.copy()
img2[np.all(img2 == (0.4, 0.4, 0.4), axis=2), :] = (0.54, 0.27, 0.27)
cv2.imshow('after: alternative solution', img2)

cv2.waitKey(0)
cv2.destroyAllWindows()

Hope that helps and clarifies!

Solution 3:

The image you have loaded is represented as a three dimensional array. The shape of it should be height, width, color channel. Ie. if it only has RGB channels (some may have RGBA etc), the matrix properties would look like height, width, 3

The rest of it should be just as you treat a normal array. You can see it as this way:

pixel = image[height][width][colorchannel]

Quang Hoang's answer has a simple solution to your problem.