Given binarized image, warpAffine() returns non-binarized image

1

I’m passing a binarized grayscale image to warpAffine(), and I’m surprised to find that it returns a NON-binarized grayscale image. In other words, I pass in black-and-white, and I get back gray. Why would rotating an image have any effect on its color values?

In case it matters, I’m using OpenCV version 4.5.something, in Python.

import cv2

img = cv2.imread("any_image.jpg", cv2.IMREAD_GRAYSCALE)
thresh, img_bin = cv2.threshold(img, 42, 255, cv2.THRESH_BINARY)   # 42 is arbitrary

print(f"Binarized image contains {cv2.countNonZero(cv2.inRange(img_bin, 1, 254))} gray values.")

height, width = img.shape[:2]
image_center = (width/2, height/2)
degrees = 5    # number of degrees doesn't matter

rotation_matrix = cv2.getRotationMatrix2D(image_center, degrees, scale=1)
    
rotated_img = cv2.warpAffine(img_bin, rotation_matrix, (width, height))

print(f"Image returned by warpAffine() contains {cv2.countNonZero(cv2.inRange(rotated_img, 1, 254))} gray values. Why?")

For my particular image file, the above code prints this:
Binarized image contains 0 gray values.
Image returned by warpAffine() contains 11631 gray values. Why?

2

the default is INTER_LINEAR. that means interpolation.

why do you expect that to not give you intermediate values?

anyway, you should threshold after the warp.

2 Likes
3

Ah, that makes sense. And now that you’ve pointed that out, I find that if I pass INTER_NEAREST instead of the default INTER_LINEAR, the result of warpAffine() is what I was hoping for (all 0’s and 255’s, no gray). Thank you for your help!