How to reshape numpy image?

Question

I have an image as a numpy array with shape (channels, height, width) how can I reshape it so that it has shape (height, width, channels)?

you can't use `numpy_image.reshape(height, width, channels)`? — farhawa, May 20 '15 at 10:24

Rob · Accepted Answer · 2015-05-20T10:42:15.973

11

I assume it would be quicker to use the built-in numpy function.

np.rollaxis(array_name,0,3).shape

edited May 20 '15 at 10:42

answered May 20 '15 at 10:31

Rob

1

don't you mean `np.rollaxis(array_name,0,3).shape`, yours will reorder it as `(height, channels, width)` – Lanting May 20 '15 at 10:36
Yes, you are correct. I got my directions confused. I will update my original post. – Rob May 20 '15 at 10:41
This function continues to be supported for backward compatibility, but you should prefer moveaxis. The moveaxis function was added in NumPy 1.11. – WurmD May 26 '22 at 13:32

Alex Riley · Answer 2 · 2015-05-20T10:51:59.637

9

You can use transpose() to choose the order of the axes. In this case you want:

array.transpose(1, 2, 0)

This creates a new view onto the original array whenever possible (no data will be copied).

edited May 20 '15 at 10:51

answered May 20 '15 at 10:46

Alex Riley

score 0 · Answer 3 · answered May 20 '15 at 10:28

0

You could do:

import numpy as np
newim = np.zeros((height, width, channels))
for x in xrange(channels):
    newim[:,:,x] = im[x,:,:]

answered May 20 '15 at 10:28

Julien Spronck

3 Answers3