How to convert black parts of an RGBA image to transparent in opencv

Question

I have an image in rgba format with (0,0,0) blacks in it, and I want to covert each of these (0,0,0) pixels to transparent. The image already has an alpha channel, and I don't know how to activate the alpha channel for a pixel. I tried setting the alpha channel value to 0, but it hasn't worked so far:

        for y in range(len(blur)):
            for x in range(len(blur[0])):
                if blur[y][x][0] == 0 and blur[y][x][1] == 0 and blur[y][x][2] == 0:
                    blur[y][x][3] = 0

Also how can I optimize this nested for loop using numpy instead?

Answer 1

It could be just displaying issue, but it hard to tell from your code sample.

cv2.imshow doesn't support transparency.
Store the result to PNG image file, and view the image using a viewer that shows transparency. (using GIMP or paint.net, it's easy to view the transparent pixels).

The code sample below does the following:

• Loads RGB image and draws black text on the image (load image as BGR).
• Converts image to BGRA.
• Sets value of alpha channel of all back pixels to zero (zero is transparent and 255 is opaque).
  The code uses logical indexing instead of using a for loop.
• Stores the BGRA image to PNG image file.

Here is the code:

import cv2
import numpy as np

# Read image in BGR format
img = cv2.imread('chelsea.png')

# Write cat in in black color
cv2.putText(img, 'cat', (img.shape[1]//2-85*len('cat'), img.shape[0]//2+100), cv2.FONT_HERSHEY_DUPLEX, 10, (0, 0, 0), 20)

# Convert from BGR to BGRA
bgra = cv2.cvtColor(img, cv2.COLOR_BGR2BGRA)

# Slice of alpha channel
alpha = bgra[:, :, 3]

# Use logical indexing to set alpha channel to 0 where BGR=0
alpha[np.all(bgra[:, :, 0:3] == (0, 0, 0), 2)] = 0

# Save bgra to PNG file
cv2.imwrite('bgra.png', bgra)

# Show image for testing
cv2.imshow('img', img)
cv2.imshow('bgra', bgra) # The letters are black, because cv2.imshow does not support transparency.
cv2.imshow('bgra[:,:,3]', bgra[:,:,3])
cv2.waitKey(0)
cv2.destroyAllWindows()

---

Result:

bgra.png file:

The site background is white, so the transparent pixels are white...