First Horsehead: Should there be less noise than this...

[feilimb]

Last week I captured 92minutes of 4minute light subs @ ISO800, in my first proper go at the Horsehead Nebula.  I was using Canon 450D (unmodified) and a Astronomik CLS filter.  I captured 16 darks of the same exposure length and 15 flats, plus bias frames.

After combining (in DSS) and processing (in PixInsight) I was delighted to see the Horsehead appearing, but a bit frustrated with how much background noise there is still (I thought this would be better with the exposures and calibration frames)..  Is the amount of noise to be expected, and the only way to get rid of it with further lights & darks to boost the signal?

I know dithering can help with all this, but I am having issues using dithering and can't seem to get it to work with my setup.  Any help greatly appreciated!

 

[MarsG76]

The primary reason for noise is the fact that it's an unmoddified camera, it is simply not sensitive enough to the Halpha (656.28nm) wavelength which the Horsehead, or Barnard 33, is very bright.

The other reason is that 23 subs is not enough data stacked to give you enough signal-noise ration for such heavy stretching which is needed when this object is captured with a unmodded DSLR.

 

You can definitely improve the noise level and horsehead intensity by increasing sub exposure length and capturing more subs... much more... in the area of 7-10 hours worth in a few nights which might be needed for a unmodded cam.

 

[Thalestris24]

Just to add - you also need to increase the numbers of calibration frames to ~30 of each. It's worth reading the dss info: http://deepskystacker.free.fr/english/theory.htm

Louise

[iwols]

thats a great start though nicely framed

[feilimb]

9 hours ago, MarsG76 said:

The primary reason for noise is the fact that it's an unmoddified camera, it is simply not sensitive enough to the Halpha (656.28nm) wavelength which the Horsehead, or Barnard 33, is very bright.

The other reason is that 23 subs is not enough data stacked to give you enough signal-noise ration for such heavy stretching which is needed when this object is captured with a unmodded DSLR.

 

You can definitely improve the noise level and horsehead intensity by increasing sub exposure length and capturing more subs... much more... in the area of 7-10 hours worth in a few nights which might be needed for a unmodded cam.

 

Thanks Mars, that makes a lot of sense now. I was just guessing up to this point but yes it did seem I was stretching things a lot more than with my previous target (M42) a few weeks ago.  So I guess the further you stretch / the fainter the signal, the more you battle with noise.  7-10 hours made me jump a bit I will certainly try to gather more light subs if I get a chance before it is too late in the season for the Horsey.

8 hours ago, Thalestris24 said:

Just to add - you also need to increase the numbers of calibration frames to ~30 of each. It's worth reading the dss info: http://deepskystacker.free.fr/english/theory.htm

Louise

Thanks Louise, I was wondering about that alright.  Even though it is all automated it seems a pain to have to wait for all those darks to complete when it is late at night, work in the morning etc.  but I guess it is worth it Thanks for the link to DSS theory, I had not read that before and will try to read through it.

[Gerry Casa Christiana]

14 hours ago, feilimb said:

Last week I captured 92minutes of 4minute light subs @ ISO800, in my first proper go at the Horsehead Nebula.  I was using Canon 450D (unmodified) and a Astronomik CLS filter.  I captured 16 darks of the same exposure length and 15 flats, plus bias frames.

After combining (in DSS) and processing (in PixInsight) I was delighted to see the Horsehead appearing, but a bit frustrated with how much background noise there is still (I thought this would be better with the exposures and calibration frames)..  Is the amount of noise to be expected, and the only way to get rid of it with further lights & darks to boost the signal?

I know dithering can help with all this, but I am having issues using dithering and can't seem to get it to work with my setup.  Any help greatly appreciated!

 

The dithering would be a solution to this also. Looks what Tony Hallas calls colour mottle. Yes in my opinion it’s too much noise. What’s wrong with dithering?

[Gerry Casa Christiana]

for example on mine last year. Not modified and no guiding but I was dithering!

[gonzostar]

Thanks for posting your effort feilimb You are giving me hope that the Horsehead is achievable next season. As i also use a canon un-modded camera  

[Galen Gilmore]

Nice! Better than mine, lol.

[John78]

Try stacking it without the dark's, take the master bias dss produces rename it and use it as the dark and then stack bias (as bias) master bias (as dark), flats and lights.  The grain/noise will be better but you will have hot pixels which dithering takes care of or you can post process away.

[dph1nm]

1 hour ago, John78 said:

take the master bias dss produces rename it and use it as the dark and then stack bias (as bias) master bias (as dark), flats and lights.

Why? I suspect the master dark has the bias subtracted already, so if you copy the  master bias to the master dark and tell DSS to use both you will end up subtracting the bias twice. I would just leave the dark out completely - DSS doesn't need it.

NigelM

[Stub Mandrel]

For the time being, try a trial of Astra Image

http://www.phasespace.com.au/download.html

This is the denoise used on your image (don't over do it or you will lose stars), still leaves some mottling... you should do better with the original data.

 

[MarsG76]

10 hours ago, feilimb said:

Thanks Mars, that makes a lot of sense now. I was just guessing up to this point but yes it did seem I was stretching things a lot more than with my previous target (M42) a few weeks ago.  So I guess the further you stretch / the fainter the signal, the more you battle with noise.  7-10 hours made me jump a bit I will certainly try to gather more light subs if I get a chance before it is too late in the season for the Horsey.

Thanks Louise, I was wondering about that alright.  Even though it is all automated it seems a pain to have to wait for all those darks to complete when it is late at night, work in the morning etc.  but I guess it is worth it Thanks for the link to DSS theory, I had not read that before and will try to read through it.

I remember I was struggling with the horse before modding my 40D... after the mod, it is as different as day and night or chalk and cheese.... It is amazing the difference a piece of bluish glass makes to astro images and amount of time needed to get signal.

 

[feilimb]

3 hours ago, Stub Mandrel said:

For the time being, try a trial of Astra Image

http://www.phasespace.com.au/download.html

This is the denoise used on your image (don't over do it or you will lose stars), still leaves some mottling... you should do better with the original data.

 

Thx Stub that looks a bit cleaner alright. I have not heard of that software before I may give it a try, but I'm keen to see if I can get a bit more out of Pixinsight first -  it seems there is more than one way to skin a cat with noise reduction. Unfortunately with the stage of the moon and the 'beast from the east' (cold snowy weather system incoming) I won't manage any more exposures for the next week. 

[feilimb]

7 hours ago, Galen Gilmore said:

Nice! Better than mine, lol.

Cheers but I really like the eye of Sauron that you captured in your image! 

[Galen Gilmore]

4 hours ago, feilimb said:

Cheers but I really like the eye of Sauron that you captured in your image! 

Ah, well it was on purpose (obviously).

HH and flame nebula were just secondary targets. 

[wxsatuser]

On 2/25/2018 at 23:57, MarsG76 said:

The primary reason for noise is the fact that it's an unmoddified camera, it is simply not sensitive enough to the Halpha (656.28nm) wavelength which the Horsehead, or Barnard 33, is very bright.

The other reason is that 23 subs is not enough data stacked to give you enough signal-noise ration for such heavy stretching which is needed when this object is captured with a unmodded DSLR.

 

You can definitely improve the noise level and horsehead intensity by increasing sub exposure length and capturing more subs... much more... in the area of 7-10 hours worth in a few nights which might be needed for a unmodded cam.

 

Agreed!

At the very least double the amount of subs and aim for a minimum of 4hours total, and get the dithering sorted.
Dithering will kill the colour noise and reveal so much more.

[wimvb]

10 hours ago, feilimb said:

but I'm keen to see if I can get a bit more out of Pixinsight first -  it seems there is more than one way to skin a cat with noise reduction.

My weapons of choice:

TGV noise reduction on L only, then MMT noise reduction on chroma only. Both on the linear image.

Btw, your image still has a gradient left. Try DBE after stacking and cropping.

I think you've managed very well on your image. Despite the heavy stretching, you've kept Alnitak under control and even split it.

[John78]

19 hours ago, dph1nm said:

Why? I suspect the master dark has the bias subtracted already, so if you copy the  master bias to the master dark and tell DSS to use both you will end up subtracting the bias twice. I would just leave the dark out completely - DSS doesn't need it.

NigelM

That was the recommended path to me from vastly more advanced imagers than I, I tried it, it works well - it is canon specific technique to reduce the noise, not having a great deal of success finding the thread on here at the minute that discussed it, you may well be right however that simply not having a dark at all would also produce the desired effect - I don't know the answer to that

With dithering and sigma clipping you negate the need for dark's entirely so you can get more time capturing signal which can only be a win.

[ollypenrice]

On 2/26/2018 at 19:00, dph1nm said:

Why? I suspect the master dark has the bias subtracted already, so if you copy the  master bias to the master dark and tell DSS to use both you will end up subtracting the bias twice. I would just leave the dark out completely - DSS doesn't need it.

NigelM

Yes, what I would try would be the master bias, renamed, as Master Dark and the original master bias as Flat Dark. This will not double-subtract the bias from the lights.

I think the image is sound for an unmodded camera and a short data set. The images of this target which the OP may have seen and liked quite possibly have ten hours or more of data - including strong Ha signal captured by a modded DSLR or through an Ha filter in a monochrome CCD. No free lunches in this game!

Olly

[feilimb]

11 hours ago, wimvb said:

My weapons of choice:

TGV noise reduction on L only, then MMT noise reduction on chroma only. Both on the linear image.

Btw, your image still has a gradient left. Try DBE after stacking and cropping.

I think you've managed very well on your image. Despite the heavy stretching, you've kept Alnitak under control and even split it.

I went back to the linear stacked version and tried playing around with TGV Denoise and MMT (after looking at some tutorials on Light Vortex Astronomy) and now have this version. Certainly the noise is better reduced, but somehow the colour went differently to my first version and I prefer the colour of the original one at top post.. ah well it's all a learning curve!

By the way here is the stacked version (142MB on WeTransfer, the link will be valid for 7 days) if anyone wants to have a go.. note however that the stacked images were FITS files, and colour channels were not aligned during stacking: https://we.tl/14BXOWtYLP

[Stub Mandrel]

5 minutes ago, feilimb said:

I prefer the colour of the original one at top post

I think this is a big improvement.

Don't despair... you have brought out the blue reflection nebulae AND picked up how the Flame is yellower than the nebulosity behind the horsehead.

[wimvb]

This is about as much as I can pull out of it, without blowing up the noise.

(click on image to see a larger version)

The noise is predominant in the red channel, because the camera is not modded and the field is filled with Ha. The red channel had a fair number of black pixels, which I removed with cosmetic correction before noise reduction. Careful stacking can maybe avoid this, but it's also a sign of the red channel being underexposed. Dithering would help to get rid of colour noise, as Gerry pointed out.

The original has a strong colour cast. Light pollution? If so, you need to gather a lot more data to combat the sky noise.

I split the image in its RGB components and did TGV denoise and MMT on each channel. Then recombined and stretched with Histogram Transform and Curve Transform. Desaturation of the red mottle with a mask, protecting the nebulae.

 

[feilimb]

7 hours ago, wimvb said:

This is about as much as I can pull out of it, without blowing up the noise.

(click on image to see a larger version)

The noise is predominant in the red channel, because the camera is not modded and the field is filled with Ha. The red channel had a fair number of black pixels, which I removed with cosmetic correction before noise reduction. Careful stacking can maybe avoid this, but it's also a sign of the red channel being underexposed. Dithering would help to get rid of colour noise, as Gerry pointed out.

The original has a strong colour cast. Light pollution? If so, you need to gather a lot more data to combat the sky noise.

I split the image in its RGB components and did TGV denoise and MMT on each channel. Then recombined and stretched with Histogram Transform and Curve Transform. Desaturation of the red mottle with a mask, protecting the nebulae.

 

@wimvb that is wonderful, thanks for taking the time to see what you can get with the data.  I really like the final result you got there, of course it does need more signal but from what is available it looks much better than my attempts.  As regards the colour cast, I guess you mean the 'cyan' cast as far as I know it is coming from the Astronomik CLS clip in filter that I used.  The next opportunity I get I think I will shoot 5 or 6 minute light frames if I can manage them.

Just curious - why did you apply the noise reduction processes separately on the different channels?  When I tried it yesterday I simply used the linear RGB composite for the noise reduction.

[wimvb]

The red channel had about 3 times more noise (standard deviation in pixinsight statistics tool), than blue or green. By seperating the channels, I could finetune the noise reduction and keep as much detail as possible in each channel. Another way to do this would be to extract the luminance, and just work on that. Then blur the colour image and use lrgb combination. I just wanted to test something new.