CODEWITHSUNDEEP
pythonnumpyopencvimage-processing
MSI
MSI

Reputation: 125

Image channel missing after resizing image with OpenCV

After resizing, the channel of the binary image is not showing. For this reason I had to use np.expand_dims method. Any solution to how I can overcome this situation, Thanks.

img = cv2.resize(img,(256, 256))
img = model.predict(np.expand_dims(img , axis=0))[0]

If I print the shape here it shows (256,256,1)

But I need to resize it for further process. If I only resize by

img = cv2.resize(img ,(720,1280))

Here the shape becomes (720,1280) # (channel number -> 1 is missing here)

If I do as below it works fine with shape (720,1280,1). But it's slowing down the frame rate as a result my processed video becoming laggy!!

img = np.expand_dims(cv2.resize(img ,(720,1280)), axis=2)

Upvotes: 3

Views: 3326

Answers (1)

aminrd
aminrd

Reputation: 5000

So what cv2.resize() does is when it receives an image with at least 2 channels, it keeps the extra band, but when it is only one channel, it drops it.

import numpy as np
import cv2

img1 = np.random.random((64, 64, 1)) * 255
img2 = np.random.random((64, 64, 3)) * 255

cv2.resize(img1.astype(np.uint8), (256, 256)).shape
# Output (256, 256)
cv2.resize(img2.astype(np.uint8), (256, 256)).shape
# Output (256, 256, 3)

I would suggest if you are always having a single band/ channel, keep using that and upon prediction on your model, pass it like this:

img = cv2.resize(img,(256, 256))
# So img.shape = (256, 256)
img = model.predict(img[None, ...])[0]

And same after that:

img = cv2.resize(img ,(720,1280))[None, ...]

Upvotes: 4

Related Questions