Problem Description:

When learning the MNIST machine learning introductory course, the attributeerror: module ‘SciPy. Misc’ has no attribute ‘toimage’ error occurs when converting the numpy array into a picture and saving it
the execution code is as follows:

# Save the first 20 images
for i in range(20):
    image_array = train_images[i]
    #Save file as the format:mnist_train_0.jpg,mnist_train_1.jpg,...
    filename = save_dir + '/mnist_train_%d.jpg' %i
    #save image_array to image
    # use scipy.misc.toimage to convert to image and save
    scipy.misc.toimage(image_array, cmin=0.0, cmax=1.0).save(filename)

Crux attribution:
I checked the scene where Python 3 uses the toimage() function on the Internet and found that this function has been deprecated. Many tutorials recommend reducing the version of the third-party SciPy to match the use of the toimage() function. I feel that this is a bit “married down” and can not fundamentally solve the problem. Moreover, the iteration of technology update is inevitable, We need to follow the trend.

Scipy.misc.toimage official latest instructions.

Solution:

Turn the problem into how to save the array as an image?

Method 1: use the .Imwrite() function of CV2 module to convert the numpy array into an image and save it. The specific codes are as follows:

cv2.imwrite(filename, image_array)  #Mutual transformation of images and numpy arrays using cv2

Method 2: use the .Fromarray() function under Image in PIL module to convert the numpy array into an image and save it. The specific codes are as follows:

from PIL import Image
Image.fromarray((image_array)).save(filename)  #Mutual transformation of images and numpy arrays using PIL
# or
Image.fromarray((image_array*255).astype('uint8'), mode='L').convert('RGB').save(filename)  # For processing color images

Method 3: use the Matplotlib module to convert the numpy array into a picture and save it. (including two methods): (1) preferred recommendation:   Use the . Imsave() function under Image in the Matplotlib module to convert the numpy array into an image for saving. The specific codes are as follows:

from matplotlib import image
image.imsave(filename,image_array,cmap='gray')  # cmap is often used to change the drawing style, such as black and white gray, emerald green virdidis

(2) Not recommended:   Use the .Savefig() function under pyplot in Matplotlib module to convert the numpy array into a picture for saving. The generated picture contains coordinate axis and border information. The specific codes are as follows:

import matplotlib.pyplot as plt
# Drawing pictures
plt.imshow(image_array,cmap='gray')
# save image
plt.savefig(filename) # In this case, the picture variable is already specified when drawing the picture, so there is no need to specify it again when saving