3D RGB Numpy 数组到图像文件导致类型错误

Question

from PIL import Image

array = np.zeros((3,64,64),'uint8')
array[0] = redToImage;
array[1] = blueToImage;
array[2] = greenToImage;

img = Image.fromarray(array)
if img.mode != 'RGB':
   img = img.convert('RGB')
img.save('testrgb.jpg')

I have redToImage , blueToImage , greenToImage and all of them is a numpy array with (64,64) size. However when I try to create image from array, it gives me that error. I really searched and tried lots of methods.

It gives this error:

***---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
~\Anaconda3\lib\site-packages\PIL\Image.py in fromarray(obj, mode)
   2514             typekey = (1, 1) + shape[2:], arr['typestr']
-> 2515             mode, rawmode = _fromarray_typemap[typekey]
   2516         except KeyError:
KeyError: ((1, 1, 64), '|u1')
During handling of the above exception, another exception occurred:
TypeError                                 Traceback (most recent call last)
<ipython-input-331-9ce9e6816b75> in <module>
      6 array[2] = greenToImage;
      7
----> 8 img = Image.fromarray(array)
      9 if img.mode != 'RGB':
     10     img = img.convert('RGB')
~\Anaconda3\lib\site-packages\PIL\Image.py in fromarray(obj, mode)
   2515             mode, rawmode = _fromarray_typemap[typekey]
   2516         except KeyError:
-> 2517             raise TypeError("Cannot handle this data type")
   2518     else:
   2519         rawmode = mode
TypeError: Cannot handle this data type***

Answer 1

The typing error says that Image.fromarray(array) cannot automatically reshape a (3, 64, 64) matrix to a (64, 64, 3) matrix. fromarray(x) expects that x will contain 3 layers or 64x64 blocks instead of 64 layers of 3x64 blocks. Changing your code to something similar as below produces the desired result (in my case a green 64x64 pixels .jpg image).

from PIL import Image
import numpy as np

array = np.zeros((64, 64, 3), 'uint8')

'''
Creating dummy versions of redToImage, greenToImage and blueToImage.
In this example, their combination denotes a completely green 64x64 pixels square.
'''
array[:, :, 0] = np.zeros((64, 64))
array[:, :, 1] = np.ones((64, 64))
array[:, :, 2] = np.zeros((64, 64))

img = Image.fromarray(array)
if img.mode != 'RGB':
    img = img.convert('RGB')
img.save('testrgb.jpg')