What if...???

[R26 oldtimer]

This is probably silly, but anyway here it goes....

What would happen if say someone intended to shoot 100 of 60sec subs of orion nebula or Andromeda, and clouds rolled in the middle of the session or just got bored and stopped for a hot beverage.

Now this poor fellow is left with only 50 subs and starts thinking of copying and renaming these 50 subs in the same folder (unaltered or with slight denoise or slight Gaussian blur) and stacking all of the 100 subs?

[henry b]

AH HA!  now the RF is kicking in and 3dB comes to mind. I would be interested to know your results if you try. Signal to noise is the question and can you reduce imaging time ?          Henry b.

[geordie85]

You'll probably end up with a worse image than with just using the 50 subs. Stacking helps improve SNR and reduce random noise. If the noise is exactly the same in half your subs it won't be reduced. Chances are, it'll be stacked and appear worse. 

At least I believe is true. I'm sure someone who is more knowledgeable than me will chime in soon enough. 

[henry b]

Ah but!    if your processing software is looking for a noise variation or change it could work, ........depends how you feed your your package?    purists are lurking around as allways.    Henry b.

[geordie85]

But there will be less variation due to noise being present in exactly the same place/ way in half of your stack. 

[henry b]

So you have tried it?    Henry b.

[R26 oldtimer]

Not yet, I thought about asking here before giving it a go. I will have a go at it (all three scenarios), although I am convinced that AP and shortcuts don't mix well!?

[adyj1]

Hmmm - I reckon, that if you could sample the same photo twice and actually make an improvement, the stacking software would be programmed to do it automatically... Many times ? 

[geordie85]

2 hours ago, adyj1 said:

Hmmm - I reckon, that if you could sample the same photo twice and actually make an improvement, the stacking software would be programmed to do it automatically... Many times ? 

And we wouldn't spend hours taking subs of the same target. We could just duplicate 1 sub as many times as we like. 

Unfortunately this is not the case. 

[ollypenrice]

This will not work, as I think we all know. 

If you take 20 measurements of the same thing as precisely as you can and average them you will get a considerably more accurate result than if you make the measurement only once. Basic science. But if you take the single measurement, interfere with it in a random way 20 different times and then average the result, you will get your original single measurement with the addition of the average of the random modifications you introduced. This will be worse than the single measurement, or at least no different from taking the original measurement, dreaming up a random change to it and applying that. It might be better but it is far more likely to be worse.

There is no point whatever in modifying part of a stack one way and another part of a stack another and combining them unless your modifications are so inept that they are best reduced by an averaging process! In this case it would be better not to make the modifications in the first place. If they are going to improve the image they should be done to the whole set. If they are not, they should not be done at all. 

The best possible data set you can take into post processing is the one with the largest number of good sub exposures properly stacked and calibrated using an algorithm appropriate to the number of subs in the set.  After that you have control over everything you do. Exercising that control rather than throwing a bunch of guesses into different subsets in the stacking process will pay dividends.

Olly

 

 

 

 

 

[JamesF]

6 minutes ago, ollypenrice said:

This will not work, as I think we all know. 

Quite.  It's probably not that hard to demonstrate rigorously using maths, but in simple terms, if you start with 50 images which are ( signal + random noise ) and stack them, you should end up with something like signal + average ( random noise ).  It may not strictly be an average of the random noise, but it will do for a simple example.  If you add in copies of those 50 images with their noise and some additional perturbation, after stacking you're going to end up with something like signal + average ( random noise + perturbation / 2 ).  Unless you got very lucky with the second perturbation therefore, it's unlikely you're going to end up any nearer the actual signal than in the first case and you may well be worse off.

James

[henry b]

well thats it then!  no point in ever trying ,             Henry b.

[wxsatuser]

5 hours ago, henry b said:

well thats it then!  no point in ever trying ,             Henry b.

Proof!
A comparison of 14x900secs Ha of the Spaghetti Nebula.
This is a very faint nebula and IMO will show why stacking the same image is a no go.

The first image is 14 stacked of one sub and the second is the full 14 subs.
No calibration, both stacked in DSS and both given an STF stretch in PI.
The histogram of the 14 same subs has noticeably more noise in PI.

14 stacked of one sub.

All 14 subs

[R26 oldtimer]

You A.P. guys are busy bees?. While I was catching some z's, you moved this ahead!

Of course I understand that you can't make something out of nothing,  so it's not a comparison between stacking 20 different subs or 20 copies of the same. It's more about if the above 14sub image of @wxsatuser, could be further enhanced by adding to the stack another 14 slightly modified copies, making a total of 28 subs.

It seems to me that stacking is not a mere adding or averaging process rather a combination of those two, but then again I am really new to this, and its a bit all Greek to me?.

[ollypenrice]

17 minutes ago, R26 oldtimer said:

 

Of course I understand that you can't make something out of nothing,  so it's not a comparison between stacking 20 different subs or 20 copies of the same. It's more about if the above 14sub image of @wxsatuser, could be further enhanced by adding to the stack another 14 slightly modified copies, making a total of 28 subs.

 

The key point is, 'What are these slight modifications and how were they decided upon?' They do not come from the sky, which is what we are trying to record, so where do they come from? They come from largely random noise reduction guesswork performed before the imager had all the information at his/her disposal. It would be absurd to work on noise reduction this way. Firstly you should get the lowest inherent noise possible by using the most appropriate algorithm on the full stack. Then you look carefully at the nature of the noise and think about how and where to reduce it using 3rd party software and/or custom intervention of your own design.  (I generally do both, but I certainly never noise-reduce an image globally because, if you do, you would needlessly blur the areas of very high signal to noise in the image and damage them.) The proposed modifying of part of the stack would do exactly this because it would be done globally.

The other way to think about stacking is this: in each sub frame the same information arrives from the sky but some of the errors in the recording are random. In some subs the noise add to the sky's recorded pixel value, in others it will subtract from the sky's recorded pixel value. The result is that you get closer to the true value each time you add a sub. Now if you have only one recording of the sky how can you do this? How can you or the stacking algorithm know whether noise has increased or reduced a true pixel value? You can't know. You have absolutely no way of knowing. You can only know by repeating the measurement in the knowledge that the thing you are measuring hasn't changed but that the errors in your measurement will have changed.

We must respect the sky as the source of our information. Any astrophographer who fails to do so is photographing his own guesswork and not the sky itself.

Olly

 

[R26 oldtimer]

14 minutes ago, ollypenrice said:

We must respect the sky as the source of our information. Any astrophographer who fails to do so is photographing his own guesswork and not the sky itself. 

I guess this is true... Whether the result would be better or worst, it would be a result of "painting" rather imaging.

[ollypenrice]

12 minutes ago, R26 oldtimer said:

I guess this is true... Whether the result would be better or worst, it would be a result of "painting" rather imaging.

...and painting blind.

Olly

[henry b]

We are all artists painting with the imformation provided or created, however....the recieved "signal" can technically be improved    given some direction, it is this that interests me resulting in cleaner data therefore less collecting time.  i am sure that camera data output could be enhanced with some effort and the support of a manufacturer but cost will allways get in the way........as allways!.   Interesting times ahead.   Henry b.

[ollypenrice]

23 minutes ago, henry b said:

We are all artists painting with the imformation provided or created, however....the recieved "signal" can technically be improved    given some direction, it is this that interests me resulting in cleaner data therefore less collecting time.  i am sure that camera data output could be enhanced with some effort and the support of a manufacturer but cost will allways get in the way........as allways!.   Interesting times ahead.   Henry b.

There's a big difference between 'provided' and 'created,' here!

Also some noise is scientifically inherent to the imaging process, notably shot noise. https://en.wikipedia.org/wiki/Shot_noise 

Olly

[Singlin]

I find it easier to just take more subs on another night.?

[henry b]

With a nice glass of wine why not  .  enjoy. !    Henry b