Gaussian/Box blur with different X and Y radius

[Splendide-Imaginarius]

What did you do?

Call ImageFilter.BoxBlur or ImageFilter.GaussianBlur with the radius parameter set to a 2-tuple containing an X and Y radius.

What did you expect to happen?

The blur operation should use different radii for the X and Y dimensions. If one of the dimensions is zero, then the blur should only be done in the other axis.

What actually happened?

This is not supported, per documentation (the radius must be a single number).

What are your OS, Python and Pillow versions?

• OS: Linux
• Python: 3.11
• Pillow: 9.5.0

Example code

# im is an image

# This works
im.filter(ImageFilter.GaussianBlur(32))

# This doesn't work
im.filter(ImageFilter.GaussianBlur((32, 16)))

Other notes

It looks like there's already internal code for doing single-dimensional box blur; making the 2-dimensional blur functions (which are exposed to Python) use different radii per dimension should be pretty easy. See this code:

Pillow/src/libImaging/BoxBlur.c

Lines 266 to 281 in a5b0256

	/* Apply blur in one dimension.
	Use imOut as a destination at first pass,
	then use imOut as a source too. */
	ImagingHorizontalBoxBlur(imOut, imIn, radius);
	for (i = 1; i < n; i++) {
	ImagingHorizontalBoxBlur(imOut, imOut, radius);
	}
	/* Transpose result for blur in another direction. */
	ImagingTranspose(imTransposed, imOut);
	/* Reuse imTransposed as a source and destination there. */
	for (i = 0; i < n; i++) {
	ImagingHorizontalBoxBlur(imTransposed, imTransposed, radius);
	}
	/* Restore original orientation. */
	ImagingTranspose(imOut, imTransposed);

I would probably be happy to send in a PR, if this is a feature that you would welcome -- let me know.

[homm]

Personally, I absolutely welcome for this feature.

[radarhere]

I've created PR #7336 to resolve this.

[Splendide-Imaginarius]

Nice, thank you for saving me the trouble @radarhere :)