The fact that it claimed to be "AI-powered" made me extremely skeptical of BlurXterminator, but now that I understand how this tool works on a technical level, I'm sold.
The guy probably just wrote some code to generate randomly varying point spread functions, made a training data set by filtering a set of known images with the PSFs, and then training a neural net on the data. It's not a trivial task, but it's also not insanely difficult.
Like, the tool was made by one guy (I think?), so it can't be too difficult. It's more that most people don't have the technical skills to do it. I probably could (given sufficient time, which is the rub).
I wonder if there's a calibration method to determine the PSF of your optical system. Like, in the same way that you take bias, dark, and flat frames to calibrate, there might be a different set of images you could take to get the PSF. 🤔
Initial instinct is to make a cover for the telescope with a hole that's smaller than the spatial resolution of your telescope and shine a light through it to make a point. That's probably harder than it sounds, and it only measures the PSF at a single point. There's probably a more intelligent and mathy method to get the PSF across the entire focal plane through. More research needed.
@thomasfuchs True, but that's probably a lot easier to estimate than whatever the diffraction PSF is for your telescope. Especially if it's a particularly weird one. I don't know. It's still an interesting thing to look into, regardless.
Add comment