Data limits without filtration?

[smr]

Hi all,

Last week we finally had a clear night, and not one but two. So I decided to set about imaging M42, I've done this for my first AP target with DSLR and camera lens at 250mm. But now have a 430mm refractor and different DSLR, and understanding a bit more about AP I wanted to get a better image than my first attempt which was 45 mins of subs. So far I have got 3 hours worth of data and if I stretch very aggressively I can just see some dust starting to show through around M42. Before stretching to reveal the dust the initial stretches show a nice amount of nebulosity in the arms around the Orion nebula etc. so I am pleased with that, but I would love to get some of the dust surrounding. I think I can see more dust generally as well, though as it's such an extreme stretch to show it, it also brings through horrendous noise. 

This may sound like a stupid question as I understand that the more data the better, SNR is improved etc. but with my DSLR being unmodded and no Ha filter being used, am I going to capture more of the dust or is it so faint that only filtration will pick most of it up? I'm thinking about getting as much data as I can if it is worth it - ie. another 3-6 hours. 

 

 

[vlaiv]

Simple rule is (and this rule applies if you don't harm your data in process of capture / calibration / stacking in some odd way):

You need x4 more time to get x2 increase in SNR (or x4 more subs if you are using equal length subs under same conditions).

This means that imaging time runs as quadratic function in relation to SNR increase. Let's say you image for 4h and want to boost your snr by x2 - this means total of 16h or 3 times more than you already imaged. But if you want SNR increase by another x2 over that - it will take you total of 64h and that is massive amount of time. At some point you will just need ridiculous amounts of time to go up by another x2.

It is much more feasible to reduce your noise in the first place, and then use above method until you run out of time budget.

That would mean - dark skies, cooled low read noise camera, ...

Btw, if you want to get the feel for SNR and what signal with particular SNR looks like - you need to fiddle around with software that can generate particular types of noise and generate your own samples which you later process. I would say that baseline for recognizable feature is SNR of about 5.

I can prepare mono fits with different levels of SNR and same "image' (I'll just write some text) - so you can play around stretching it to get the feel for different levels of SNR. This will help you judge level of noise in your image "by eye" to see if another x3 amount of time is going to be significant in increasing quality of your image. Just gonna grab a bite and then I'll generate fits and post it here.

[vlaiv]

Here it is:

noise_table.fits

Image is made up from 5x5 same segments polluted with noise to produce different levels of SNR

First four rows contain

1st row: SNR 1,2,3,4,5 (from left to right)

2nd row: 6-10

3rd row: 11-15

4th row: 16-20

Last row contains higher SNR values: 25,50,75,100 and 125

Here is screen shot of unstretched sub loaded in gimp:

And here is basic stretch - first two rows (SNR 1-5 and 6-10)

Have a play with this to see what different levels of SNR look like and compare to your data - this will tell you if there will be significant improvement if you double SNR for example

(btw general formula for SNR increase would be square root of time, so if you have for example one hour and you add two more hours for total of three - you will have SNR increase of ~1.7 - being square root of 3, so if your SNR was for example around 4 - it will rise to around 6.8)

And another note - look at worst affected areas of your image to get lowest SNR that needs to be addressed (like dust in background).

 

[ollypenrice]

The area outlying the bright parts of M42 is often imaged in Ha where it does reveal strong structures. However, these structures also appear strongly in broadband (natural colour) as brown dust. So more data is certainly worth it. No question whatever.

When stretching to reveal faint signal the trick is to stretch the image till the darkest parts hit the noise level you find acceptable. (You can gently noise reduce them and stretch a bit more till you feel the NR is getting intrusive.) At this point there is no point in stretching the darkest stuff any more but if you pin the curve at that point and put a fixing point below that, you may be able to stretch a bit more just above it. This will enhance contrast in the dusty parts. However, I'd try to avoid the negative curve we see in your curves window above. The bottom part of the curve should remain a straight line and the lift should come directly off that. The negative, drooping curve will reduce contrast between all points along it, which is the precise opposite of what you're looking for.

Olly

 

 

[vlaiv]

13 minutes ago, ollypenrice said:

The area outlying the bright parts of M42 is often imaged in Ha where it does reveal strong structures. However, these structures also appear strongly in broadband (natural colour) as brown dust. So more data is certainly worth it. No question whatever.

When stretching to reveal faint signal the trick is to stretch the image till the darkest parts hit the noise level you find acceptable. (You can gently noise reduce them and stretch a bit more till you feel the NR is getting intrusive.) At this point there is no point in stretching the darkest stuff any more but if you pin the curve at that point and put a fixing point below that, you may be able to stretch a bit more just above it. This will enhance contrast in the dusty parts. However, I'd try to avoid the negative curve we see in your curves window above. The bottom part of the curve should remain a straight line and the lift should come directly off that. The negative, drooping curve will reduce contrast between all points along it, which is the precise opposite of what you're looking for.

Olly

 

 

Just a quick question regarding "negative" curve.

If it's contained to noise floor region would it not decrease contrast between noisy pixels making them more smooth?

[ollypenrice]

Just now, vlaiv said:

Just a quick question regarding "negative" curve.

If it's contained to noise floor region would it not decrease contrast between noisy pixels making them more smooth?

Yes but at a cost to contrast in the real signal. What would be the point?

What I sometimes do (because I find one of my cameras generates noise in the form of overly dark pixels) is pin the curve at the brightest point of genuinely dark sky, fix above that and stretch below it.

Before:

After: