OpenCV：如何检测像素颜色变化

Question

I have succeeded to colorize the area I want of a given picture created using

numpy (`img = np.zeros((512,512,3), np.uint8)`).

I display the picture using OpenCV

cv2.imshow()

After colorization using the mouse cursor, I save the picture.

How can I detect that the color of a given pixel of my image has been modified?

Answer 1

In general, comparing two arrays can be done with the usual == , < , != , etc operators. The comparison returns a boolean (True/False) array:

import numpy as np

x = np.array([0, 1, 2, 3, 4])
y = np.array([9, 1, 2, 3, 7])

arrays_equal = x == y

arrays_equal will be a boolean array, True where they're equal and False where they're not:

array([False,  True,  True,  True, False], dtype=bool)

---

However, there's an additional caveat because you're working with image data. What you probably want to get in the end is a 2D array of where any color has changed, but you're comparing two 3D arrays , so you'll get a 3D boolean array as output.

For example:

im = np.zeros((5,5,3), dtype=np.uint8)
im2 = im.copy()

# Change a pixel in the blue band:
im2[0,0,2] = 255

# The transpose here is just so that the bands are printed individually 
print (im == im2).T

This will yield:

[[[ True  True  True  True  True]
  [ True  True  True  True  True]
  [ True  True  True  True  True]
  [ True  True  True  True  True]
  [ True  True  True  True  True]]

 [[ True  True  True  True  True]
  [ True  True  True  True  True]
  [ True  True  True  True  True]
  [ True  True  True  True  True]
  [ True  True  True  True  True]]

 [[False  True  True  True  True]
  [ True  True  True  True  True]
  [ True  True  True  True  True]
  [ True  True  True  True  True]
  [ True  True  True  True  True]]]

Of course, what you probably wanted was something more like the last band.

In this case, you want to use np.all to "reduce" things down and get a 2D array of where any color in any pixel is different.

To do this, we'll use the axis kwarg to np.all to specify that the comparison should be done along the last axis ( -1 or 2 are equivalent in this case: -1 just means "last"):

np.all(im == im2, axis=-1)

Which yields:

array([[False,  True,  True,  True,  True],
       [ True,  True,  True,  True,  True],
       [ True,  True,  True,  True,  True],
       [ True,  True,  True,  True,  True],
       [ True,  True,  True,  True,  True]], dtype=bool)

Also note that if you needed to "flip" this array, you could either use the != operator with np.any instead of np.all or you can invert the result using the ~ (logical not in numpy) operator. Eg opposite = ~boolean_array .