Saving a Numpy array as an image

Pictures with Python: Meet the libraries

While the fast answer gets you to the picnic, let's unpack the basket for some more options.

Playing with OpenCV

OpenCV is not just library—it's a full-blown Swiss Army knife for image processing. Use cv2.imwrite() like this:

import cv2
import numpy as np

# Look at the colors dance!
array = (np.random.rand(100, 100, 3) * 255).astype(np.uint8)
cv2.imwrite('PsychedelicJPG.jpg', array) 


OpenCV got moves—it knows how to handle different formats and encoding options.

Pillow: More than just sleep

Pillow, an improvement over PIL, knows how to work with images and is as fresh as a morning croissant! Here you go:

from PIL import Image
import numpy as np

# More random values to keep things spicy!
array = np.random.rand(100, 100, 3) * 255
image = Image.fromarray(array.astype(np.uint8))
image.save('pic_with_pillow.png')


Pillow favours the brave—it supports a wide range of image formats.

Manual PNG creation: Control freak's delight

For the control freak in you, manually write a PNG file using zlib and struct:

import zlib
import struct
import numpy as np

array = np.random.rand(100, 100).astype(np.uint8)
# Here is where we get our hands dirty. Very. Dirty.
# ...implement PNG creation process...


But beware, with great power comes great responsibility—here be binary dragons!

Bulletproof your image-saving

Making grayscale great again

Dealing with monochrome? Add a cmap when using matplotlib:

# Even in black and white, we shine!
plt.imshow(array, cmap="gray")
plt.imsave("monochrome_matters.png", array, cmap="gray")


Transparency, done right

Messing with RGBA arrays? Remember, the fourth channel is the alpha—a.k.a., the ghost that makes your images transparent:

# Like sneaking bulgogi into a vegetarian's taco...nobody will notice
rgba_array = np.random.rand(100, 100, 4) * 255
imageio.imwrite('being_hauntingly_transparent.png', rgba_array.astype(np.uint8))


Scaling: Size matters

Scale your arrays for a consistent look. It's like having an image stylist for your arrays:

#We're always in style with consistent brightness and contrast!
array_normalized = (array - array.min()) / (array.max() - array.min()) * 255
image = Image.fromarray(array_normalized.astype(np.uint8))
image.save('image_scaling_vogue.png')


In-memory image processing

Why only use files when you can use memory efficiently? Write directly to a file-like object with matplotlib:

import io

# Ah, the things we can do with bytes!
buffer = io.BytesIO()
plt.imsave(buffer, array, format='png')


Before reading, give your buffer a quick rewind:

# T-minus 3...2...1...blast off! 
buffer.seek(0)
