PIL vs OpenCV

As I’m standing on the precipice of doing a bunch of image processing/classification, it occurs to me that I don’t know a whole lot about the available tools and packages for working with images. The following is a look at the two more-popular libraries.

PIL and cv2 both support general image processing, such as:

  • Conversion between image types
  • Image transformation
  • Image filtering

PIL (Pillow)

The Python Image Library

  • Easy to use
  • Lightweight

Use when you want to cut and resize images, or do simple manipulation.

Installing

Although you import the library as PIL, you have to install it using

pip install Pillow

Use

from PIL import Image

im = Image.open('images/daisy.jpg')
im

png

Documentation

Handbook here

cv2 (Open CV)

Used for Computer Vision, hence is a much more robust package.

  • Loaded with algorithms well-suited for data science and vision-based robotics.

Although, it bears a warning that the library, annoyingly, defaults to reading/writing everything as BGR, instead of the intuitive RGB, because “reasons.” Here’s an excellent post trying to explain why.

Installing

Not to be outdone, getting ahold of cv2 is even more cryptic than PIL.

pip install opencv-python

Use

Whereas PIL reads an image to an Image, cv2 cuts right to the chase and stores it as a np.array

import cv2

im = cv2.imread('images/daisy.jpg')
im.shape
(533, 400, 3)

Requiring us to lean on matplotlib to inspect the image

%pylab inline

plt.imshow(im)
plt.axis('off');
Populating the interactive namespace from numpy and matplotlib

png

Documentation

Open CV tutorials

Benchmarking

This kaggle post illustrates that OpenCV blows Pillow out of the water when performing multiple, repeated transformations over multiple files.