In image processing, alpha channel is usually used to control the transparency of an image. In itself, the alpha channel is just a gray level image1. In this post, I want to talk about how to add an alpha channel to RGB image and show it using the Pillow package.
Suppose that we have the following image:
we can add an alpha channel to it using the Image.putalpha() method:
from PIL import Image im = Image.open("owl.jpg") alpha = Image.new("L", im.size, 50) im.putalpha(alpha) im.show()
You might expect to see a new image with opaque color, right? But what shows up is nothing different from the original image? Why? Do we use the
putalpha() method wrongly?
It turns out this has something to do with how image is shown using the
show() method on different platforms. From the documentation of
Displays this image. This method is mainly intended for debugging purposes. On Unix platforms, this method saves the image to a temporary PPM file, and calls either the xv utility or the display utility, depending on which one can be found. On macOS, this method saves the image to a temporary BMP file, and opens it with the native Preview application. On Windows, it saves the image to a temporary BMP file, and uses the standard BMP display utility to show it (usually Paint).
So on Windows, the new image is converted to BMP format and shown using the system default image viewer. During the process of format conversion, the alpha transparency is apparently lost2. That is why you see no difference between and new image and original image.
To fix this issue, we can patch the Pillow code to use PNG format as default. First, we need to find the root of Pillow package:
import PIL print(PIL.__path__)
On my system, the output is:
Go to this directory and open the file
ImageShow.py. I add the following code after the line
class WindowsPNGViewer(Viewer): format = "PNG" def get_command(self, file, **options): return ('start "Pillow" /WAIT "%s" ' '&& ping -n 2 127.0.0.1 >NUL ' '&& del /f "%s"' % (file, file)) register(WindowsPNGViewer, -1)
After that, I can show the image with alpha channel correctly.
License CC BY-NC-ND 4.0