Python PIL thumbnail with Image.ANTIALIAS. How to preserve quality

5
votes

I'm trying to resize images while keeping aspect ratio. I use PIL's thumbnail method for this. I use Image.ANTIALIAS filter.

You can check my code here:

image = Image.open(design.design.path)
format = image.format
image = ImageOps.mirror(image)
new_size = (241, 241)
image.thumbnail(new_size, Image.ANTIALIAS)
image.save(response, format)

This code works perfectly however quality is lost after thumbnail. I can see it by zooming in on the saved image. I can see pixels at the corners of the image while I don't on the original image. This is seen even better when I print out resized image.

You can check out sample images here: http://imgur.com/a/ifZoU

Please tell me if you need anything else

2
pythonimagepython-imaging-library
are you making the images smaller or bigger? either way i'd recommend using the resize method rather than the thumbnail method – James Kent
@JamesKent I'm making them smaller. I'm using thumbnail instead of resize to maintain aspect ratio. Resize does not preserve aspect ratio. – Marijus
Why would resize not preserve aspect ratio? You hardcode the aspect ratio here to 1:1, and that would also be the aspect ratio if you would use resize. – physicalattraction
@physicalattraction yes in this case but this is a demo code and in real application there's a few different sizes I want to resize to which are not squares. – Marijus
Still, in that case resize keeps the aspect ratio you give it, which can be deduced before you resize it. – physicalattraction

2 Answers

2
votes

Image.save has options for you:

img.save(fpath, 'image/png', quality=100, optimize=True)
1
votes

If you are making thumbnails of a pixel-based image, of course you will loose quality. Re-sizing this kind of image (as opposed to vector images) simply throw information away - there is no way to recover it.

IF you need to view the image in full size, you have to preserve the original image, and keep the scaled-down version to be shown only where it is needed - use the original for everything else.

https://en.wikipedia.org/wiki/Raster_graphics