Hacker News Reader

Comment by rbritton | Hacker News Reader

Zoomed out, those photos aren’t terrible, but if you click to view a higher resolution version, you can see the softness created by the noise removal algorithm. Based on my experience, the image quality is comparable to DSLRs of about ten years ago. Noise removal is a bandaid. In astrophotography, it will virtually always look softer than reality actually is, and it will also have removed actual stellar objects. There is no substitute for a better, lower noise sensor, but you work with what you have.

Sensor size is inversely correlated with the noise level. Smaller sensors have more noise where the least noise is on full frame (i.e., 35mm) or larger sensors. Phones have small sensors.

View on HN

This not just simple noise removal but combining multiple photos with the noise appearing in different places (aside from the hot pixels). So, it's a strategy to get much of the same effects of a really long eposure with much less noise than you'd expect based on the tiny sensor because it averages out over multiple shots. In principle of course possible to do with a DSLR as well. E.g. Hugin might be able to do this.

Of course regular noise cancellation and other very lossy processing still kicks in after that (which may explain the blurry result). It would be interesting to look at the raw image produced by this.

I use open camera on my cheap Nokia 7 plus (which uses two cameras) and have been getting OK-ish results in Darktable. The dng file you get combines information from both sensors. One of them is black and white so these look really flat until you fix it in post processing. The raw photos have lots of noise (as you would expect) but noise filtering is pretty effective.

I imagine for this it would produce a dng with information from the different stills combined but none of the other post processing (except maybe hot pixel removal).

Yes you can do it with DSLR aswell, similar discussion back in HN[0]

[0]https://news.ycombinator.com/item?id=18452717

that's exactly what is amazing about this tech - 10 years ago you'd have to carry a backpack, now the same quality pictures can be taken with a multi-purpose device which fits in your pocket. nothing to complain about if you ask me!

I still get better photos with my almost 10 year old camera though.

UPD: oops, XT-1 is just 5 years old, actually. I guess I have to take my words back, sort of.

Still, the point is the same - as long as you make a shot for your Instagram account - your smartphone is alright. For anything bigger you still need a camera with decent lens.

Meh, I used to take better photos with a disposable pocket camera where you take off the shutter to allow for long exposure. But it’s a step in the right direction. Ultimately you would indeed need to filter a lot of thermic noise to allow for longer exposure on a CCD. (I started digital Astro photo years ago with an ST6 that needed to be cool down to -60C using a Pelletier module)

Not exactly, noise level is inversely correlated with the photosensitive element size. The smaller they are, the more noise you have. For example an 8MPX APC sensor should have less noise than a 40MPX full frame sensor.

This is an oft-repeated myth.

Per-pixel noise depends indeed on the pixel size, but image-scale noise is almost completely independent of the pixel size.

That's because while a large pixel has less noise, it appears larger in the output and so the noise is correspondingly more visible. A small pixel is noisier, but smaller, so the net noise is the same for the same sensor size.

This and sensor technology ie. MFT sized 12MP older sensors perform worse results than more recent 16MP MFT sensors which performs worse results than even more recent 20MP MFT sensors.

Reply on news.ycombinator.com