It's lossy, but not destructive, and a 'sharpen' operation is technically the same as blur but in reverse. So you won't end up pixel-perfect after doing an 'unblur' but you will be able to make out more than you could before.

If you know anything about the probability distribution of likely inputs, it's even easier to reverse with minimal loss.

Eg knowing that the input was black text on white background or a natural image (instead of eg white noise) helps a lot.

Also if you have multiple pixelated/blurry images that helps you can reconstruct it more easily, e.g. if different newspapers print pixelated picture of the "suspect" you can reconstruct it pretty accurately.

Machine learning can also do a surprising good job of it, especially if you know what the target is (e.g. a face) https://www.vox.com/future-perfect/2019/9/4/20848008/ai-mach...

Sample code: https://gist.github.com/JonathanFly/80b669a72bf624d17b56a1cf...

> Machine learning can also do a surprising good job of it, especially if you know what the target is (e.g. a face)

Yes. Though that's just a corollary of doing better when you know something about the probability distribution of inputs.

(But a very useful and practical corollary. My formulation didn't give any hint how you might make use of that knowledge of the distribution.)