Three Dollar Quill

I bit the bullet and implemented matrix-based error diffusion. Notes so far:

• When I do Floyd-Steinberg diffusion, I don't get what Pillow does when it does Floyd-Steinberg diffusion.
• Sometimes what Pillow does is better in terms of detail, and sometimes it isn't.
• Sometime Floyd-Steinberg is good, sometimes Jarvis-Judice-Ninke is good, and I don't think I've yet hit a situation where Atkinson was the standout. I probably need to try a greater variety of images.
• At least the way I write it, matrix-based error diffusion is incredibly slow.

Looks like I should work on a few categories of thing later:

• Cleaning this up and reworking the relevant interfaces.
• Finally generating blue noise in the format that I want.
• Some quick experiments showed that simply increasing the input image brightness (by quite a lot, in the case of Atkinson dithering) can make things come out a lot clearer. Perhaps there's some way to automate the offset determination?
• I've still got a few more average implementations I want to try, but, like, realistically speaking, they're not likely to accomplish anything groundbreaking. Nothing with the aesthetic punch of "Atkinson dithering but first I added one eighth of full brightness are you kidding me?" (I have no idea what this corresponds to mathematically, or how much it messes with the theoretical properties of Atkinson dithering.)

Anyway, we'll see if I'm up for any of that tomorrow.

Good night.