Hacker News Reader

Comment by ajkjk | Hacker News Reader

Would it be possible for them to algorithmically remove the lens artifacts? I suppose just "anything with hexagonal symmetry gets deleted". They're kinda pretty but it's sad to think that this isn't what it would actually look like if you had really good infrared eyes.

View on HN

I've looked a little at this in the context of under display cameras, which introduce similar artifacts. They could remove these artifacts and generate realistic looking images that are physically plausible, but the content in regions where the artifacts are present are under determined and require some "guessing" or artistic expression. Probably better for NASA to keep things closer to the captured data in their primary releases.

Some astro photographers use StarNet[1] to remove stars altogether. This allows you to blend them back in[2], potentially after additional processing.

As the name suggests, StarNet uses neural network for the job[3], and at least yet hasn't been trained on Webb data AFAIK, so might not do a good job there.

There also seems to be a similar tool with the same name over on github[4], but seems to be a separate effort.

[1]: https://www.starnetastro.com/

[2]: https://astrobackyard.com/starnet-astrophotography/

[3]: https://www.cloudynights.com/topic/808556-starnet-v2/

[4]: https://github.com/nekitmm/starnet

Looks incredible, thanks for sharing! Hope they can remove stars from Webb’s Pillars of Creation at some point.

Interesting- If you look closely at the super bright objects, there are both hexagonally symmetric artifacts and cardinal up/down/left/right artifacts. I’d love to hear from someone with knowledge of how these arise and what optical tradeoffs prevent designs from eliminating them.

It's possible through deconvolution with a model of the "point spread function" aka PSF. There are techniques from radio astronomy (where interferometry necessitates some kind of Fourier basis model and deconvolution/imaging procedure) which has been extended to the optical domain.

See for example: https://twitter.com/NGC3314/status/1582442367482634242?t=Eaa...

Isn't the PSF intensity-dependent? Can deconvolve high bit-depth data with a single filter? If not, aren't you limited by the distribution of intensities of the point sources present?

You can read about their formation on several publications. For instance: https://bigthink.com/starts-with-a-bang/james-webb-spikes/

There are 2 causes of diffraction spikes:

1. The hexagonal mirrors. Trade off is larger mirrors, and the ability to fold the telescope. 2. The struts holding the secondary mirror. Basically can't get rid of this without a very different telescope design.

Causes of both are the best design for getting a lot of light into a space telescope's sensors. In practice has less impact on the science than you might think. The diffraction spikes are much fainter than the main image. It's only bright foreground stars - which the James Webb isn't trying to image anyways - that cause visible spikes.

Rotate the camera and bit and take another photo?

>Rotate the camera a bit

Not that easy if "the camera" is JWST.

What we need is 2 JWST that can compensate for each other.

Why build one when you can have two at twice the price?

The price is mostly deploying, not building.

As I understand it, they aren't exactly lens artifacts, they're stars in the foreground. The telescope can't see past them and we don't know what's there. To me, it would defeat the observational mission of the JWST to fill it in with some pixels that were similar to the rest of the background. However, I also think you raise a good point, since all these images are constructed from non-visible infra-red light and many decisions are made in that conversion process.

They're not talking about the stars, but the hexagonal spikes sticking out of the stars. Those are artifacts of the hexagonal lenses that the telescope uses.

I think you're both right to a degree? They are lens artifacts and we can't just remove them because we don't have the data of what's behind the spikes, as that is what the sensor can see. We'd be filling them in with black or our best guess based on neighboring pixels, either way it wouldn't be what one would "really" see. Unless we composited images from another source, then that'd come closer to accurate.

That seems right. For popular consumption it'd probably be nice to have a black-filled version or something, whatever is least visually off-putting, but for science that seems much, much worse.

No, the hexagonal spikes are artifacts of the arms holding the secondary mirror in place.

If you removed them wouldn't you just see black areas where they were?

Dunno. I'm guessing that you could remove them and see stars underneath them, because, after all, other parts of the telescope can see that patch of sky (the arms don't _entirely_ occlude it.. just from some angles). But I don't really know what I'm talking about.

Reply on news.ycombinator.com