RGBA conversion breaks edge detection

[BernardZhao]

What did you do?

I am having an issue with Pillow's edge detection, specifically after I convert my source image to RGBA. For context, I open a .jpg file, and edge detection works fine on the image. However, after conversion the detected edges are all blank. Am I doing something wrong that involves the new alpha channel? See the code below to replicate:

What did you expect to happen?

I expect the edges of the images to be detected, producing an image like this:

What actually happened?

Instead, a blank image is generated:

What are your OS, Python and Pillow versions?

• OS: Mac OSX Catalina
• Python: 3.7.5
• Pillow: 6.2.1

from PIL import Image, ImageFilter
image = Image.open("examples/image.jpg")
image.filter(ImageFilter.FIND_EDGES).show()
image = image.convert('RGBA')
image.filter(ImageFilter.FIND_EDGES).show()

I have also tried this with other ImageFilters, and they seem to work fine.

[radarhere]

My interpretation of what you're seeing is that Pillow is finding the edges of all channels, including the alpha channel. Since your image has a constant value for the alpha channel throughout - opaque - the edges of the alpha channel are the edges of the image - a box.

So if I run the following code on a test image that actually has varying alpha values -

from PIL import Image, ImageFilter
im = Image.open("Tests/images/pil123rgba.png")
im.filter(ImageFilter.FIND_EDGES).save("pil123rgba_edges.png")

[homm]

Instead, a blank image is generated:

It's not blank, it's transparent (have zero in the alpha channel). If you want to preserve the alpha channel, you can use alpha = im.getchannel('A') and im.putalpha(alpha) methods.