Where should I focus to improve

[Penguin]

Hi Guys, just looking for some general help with improving my images I'm not sure what to concentrate on to get the best bang for my buck. Here are my latest couple of images:

Both of these are taken with an unmodified Canon 450D through a Skywatcher Explorer 150P on an HEQ5 Pro which is on a permanent pier. I have no guide-scope or filters and just take colour subs. I am stacking in DSS and processing in PixInsight, blindly following a workflow I was shown a couple of years ago. The PixInsight workflow goes:

1. Background Neutralisation
2. DBE
3. Colour Calibration
4. Histogram Transformation
5. HDR Multiscale Transform
6. ACDNR, followed by another bit of histogram transformation
7. Curves Transformation (I possibly overdo this a bit but I like the colours!)
8. SCNR

Both these images were with 2 minute subs. The Flame/Horsehead was 2 hours-worth (with 15 darks and no flats) and M81/M82 was 4 hours (with 15 darks and 15 flats).

What do you think are my worst offences? I can think of lots of things I could improve but I don't know what to attack first:

• Better polar alignment
• Guiding (I do have an 80mm Skywatcher refractor that I can piggy-back but I don't have a guide camera)
• Bias frames (I believe I could just take one master set rather than taking them every session. Would they have much of an effect?)
• Using filters (maybe a light pollution filter?)
• Coma Corrector
• Better use of DSS
• Better use of PixInsight

[bottletopburly]

Do you use a bahtinov mask at all?

[Penguin]

Yes, I focus with a bahtinov mask, but I always end up with one of the axis spikes diverging into two, as can be seen with Alnitak on the Horsehead/Flame image and I don't understand what is causing that. I tend to focus at the scope since I don't have an auto-focuser so would have to keep going back and forth between scope and computer. Otherwise, I would post the focusing images for you.

[alan4908]

21 hours ago, Penguin said:

Hi Guys, just looking for some general help with improving my images I'm not sure what to concentrate on to get the best bang for my buck.

Looking at the images I'd say that your main area to concentrate on would be on noise reduction since both images exhibit a high degree of colour noise and the so called "colour blotch" effect.  Whilst you can reduce the effect of these by using TGVDenoise in Pixinsight I'd also suggest you concentrate on calibrating your data correctly with flats, darks and bias frames - incorrect calibration can result in excessive noise. 

I'd also suggest that you look at the image after using DBE or Pixinsight's CC scheme - often you will see a green cast than can be eliminated by use of SCNR green. Have a go at applying SCNR green on the first image !

Your focus and tracking, whilst not perfect, are secondary issues in my opinion.

Hope this helps.

Alan    

 

 

[Penguin]

Thanks Alan, there's some useful (and free!) things I can look into there.

I have not tried TGVDenoise so I will investigate that. I normally do SCNR Green as the last step in my workflow but having just tried it on the Flame/Horsehead image, I can see that it is reducing the green so maybe I had missed that step.

In terms of calibration frames

• I normally take 15 darks at the end of my session. It's a bit annoying having to spend another 1/2 hour taking 'blank' shots before packing up when its already 12:30am but I do understand the benefit. Is 15 frames/30 minutes long enough for the darks?
• I also take around 15 flats at the end of the session using this panel, which has several sheets of photo paper in front of it to allow the camera to take a long enough exposure that the LED refresh doesn't mess up the image, and is housed in a makeshift frame. I have also attached a single flat from the most recent session. Again, is 15 enough? Is there a better way to take them?
• Bias is easy and I gather I can prepare a set once and use them multiple times, I have just been lazy. I will prepare a set now and might try reprocessing a few of my existing images.

 

[michael8554]

Hi Penguin

Save yourself a heap of time by ditching the Darks and taking Bias (fastest shutter setting on the 450D) and loading those as Darks in DSS.

But you must Dither between shots.

Michael

[Penguin]

Hi Michael, that would be fabulous to save time or get an extra few lights. Can you explain how using bias as dark frames works? I thought a primary purpose of darks was to get the thermal signal with the light-frame exposure at the temperature the lights were taken at so that it could be subtracted from the lights. How do we achieve that without darks? Also, DSS complains if the darks are not the same exposure as the lights. Will it still work?

Also, what do you mean by "you must Dither between shots"?

[michael8554]

First, an error in my first post.

Should be fastest shutter setting on the 450D for Bias, not AV - that setting is for the Flats.

How does it work? Dunno, it was a suggestion by Ollie the Imaging Supremo.

DSS doesn't mind the difference in exposures.

Does it work? Yes, I can't see any difference with/without Darks.

Dithering averages out the DSLR noise pattern.

Michael

[alan4908]

1 hour ago, Penguin said:

In terms of calibration frames

• I normally take 15 darks at the end of my session. It's a bit annoying having to spend another 1/2 hour taking 'blank' shots before packing up when its already 12:30am but I do understand the benefit. Is 15 frames/30 minutes long enough for the darks?
• I also take around 15 flats at the end of the session using this panel, which has several sheets of photo paper in front of it to allow the camera to take a long enough exposure that the LED refresh doesn't mess up the image, and is housed in a makeshift frame. I have also attached a single flat from the most recent session. Again, is 15 enough? Is there a better way to take them?
• Bias is easy and I gather I can prepare a set once and use them multiple times, I have just been lazy. I will prepare a set now and might try reprocessing a few of my existing images.

Darks: a dark frame should be at the same temperature and the same duration as the lights, since thermal noise is linearly proportional to the exposure duration, some calibration software can automatically adjust the dark frames to match the duration of the light frames. You appear to be taking 15 DARKS each of 2mins (to match the lights) immediately after you have taken the lights, since your camera is not cooled. This should be OK. I'd suggest you stack these using some form of non-linear rejection eg sigma clipping which will reject any cosmic rays hits - don't use average or mean stacking since one cosmic ray strike will significantly impact the result. With very low noise and cooled CCD camera's you may not need to take DARK frames, however, since you have a non-cooled DSLR, I'd suggest you follow the dark frame subtraction procedure.

Bias: these capture the camera's bias signal and read noise - again use a non linear rejection when stacking to eliminate cosmic rays - I normally create 100 bias frames. The stacked bias frame (call it Master BIAS) is subtracted from the stacked flat and stacked dark frames to create the Master Flat and Master Dark frames. I'm assuming here that you don't need to take so called Flat Darks which are a dark frame for the flats, normally subtracting a Bias frame for the flat is a good approximation.

Flats: I use an EL panel since it has a relatively uniform spectrum, so provided that your LED panel also has a uniform spectrum that should be OK. I put your flat through CCDInspector which shows that you are picking up vignetting, which is good (see below). However, what I couldn't see is any dust donuts etc - which is odd. Normally, I exposure my flats with a duration such that I'm 30 to 40% of my camera's saturation limit. Since you have a colour camera you also need to apply a boxcar filter to your individual flats before stacking, this will destroy the bayer matrix. For flats, to create the stack, you can use a linear rejection algorithm (eg median). 15 FLATS should be OK.

So here, a calibrated light frame = (Individual light frame - Master Bias - Master Dark)/Master Flat

Your flat:

 

Alan

[ollypenrice]

The real average temperature of your shots is very hard to ascertain, as is the real average temperature of your darks - yet they need to match. This is the problem. According to Tony Hallas who (unlike me) does a lot of DSLR imaging, you will do better by using a large scale (12 pixel) dither and using bias-as-dark. The large dither moves the hot pixels around and also combats what he calls 'colour mottle' - and you certainly have that in the HH image. For what it's worth, like many others using set point cooled CCD, I don't use darks either. I find they can be more trouble than they're woth.

On another tack, while I do use PI for some post processing tasks I also think it's a complicated palaver of a programme to use. It is mask rather than layer based so you somehow need it to make exactly the mask you want. In a layers based programme you don't need to worry about that. You modify a bottom layer and let it through where you want to by erasing the original top layer. This is heresy to the theologians of the Spanish Inquisition but I think it's nice and simple.

Olly

[Penguin]

Thanks guys, there's a lot to think about here, I shall take it away and cogitate.

[wimvb]

Regarding processing: when you use DBE, you put samples too close to DSO's and stars. The dark halo around the main targets is a tell-tale sign.

Try this workflow
 

1. Use Dynamic Crop to crop any stacking artefacts. Your image should have clean edges.

2. DBE, use no more than 9 samples: 1 in each corner, 1 alongside each edge, 1 somewhere in the centre, but away from any DSO or star halo. Increase the sample size to about 25 - 35 pixels. Make sure you don't cover any (bright star). Increase tolerance until all samples are accepted. First do a test run where you examine the background model. This should be smooth. Apply STF to the background model. If this looks like there is vignetting, use Division as correction method, otherwise use Subtraction. Then do the real background correction. DBE alters the background, so you should do it before background neutralisation.

3. Background neutralisation, use the largest preview you can fit over a patch of clear background, as reference.

4. Colour Calibration, use the same preview for background reference, and the entire image, or a large preview that contains an entire galaxy, as white reference. If using a galaxy as white reference, deselect structure detection.

These four steps should get you a clean image that you can continue working on.

 

Good luck.

[Penguin]

Thank you so much guys, I am learning again!

I took a load of bias frames at ISO 1600 (matching the lights/darks) and 400 (matching the flats) and re-stacked the HH image in DSS, which warned about differing ISO settings but seemed to apply them correctly. Then I applied Wim's workflow which was a slight re-ordering of steps and very different with DBE: I had previously been told to use 30 samples per row, which often seems to extract a very mottled background and I suspect may be a major cause of the problems I have been having. For colour calibration, I have never previously used the entire image for white reference, nor used structure detection.

Here are the results, it's much cleaner but has also lost dome detail and definition, particularly noticeable in the Horsehead itself. Possibly some aspects of my previous processing were better than this attempt:

 

[wimvb]

Much better, in my opinion.

Still a gradient towards the bottom (especially bottom right). You can try applying DBE twice with tweaked settings. But because of the extended nebulosity and bright star halos, keep the number of samples low.

 

[alan4908]

Yes, that is a considerable improvement.

Alan

[Adreneline]

On 28/12/2017 at 16:46, Penguin said:

I am stacking in DSS and processing in PixInsight, blindly following a workflow I was shown a couple of years ago.

I too started off with pretty much the same approach but have modified my methods over the past 12 months and you may find this of interest:

• I use the BatchProcessor in PI to calibrate, register and stack my images; I've tried doing it the long way and found no discernable benefit.
• I then dynamic crop the master and apply ABE twice, first with subtraction and then with division; again I've tried using DBE but in most cases have found ABE made what I considered to be a better job of it.
• Background Neutralisation followed by Colour Calibration as described by Wim followed by SCNR
• I then create a mask using Channel Extraction CIEL (no 'a' or 'b') and apply inverted to the image.
• I then use MLT to reduce the noise on the first four layers as described in a Light Vortex tutorial but I adjust the values whilst monitoring the preview to try to make sure I don't over do it - always very tempting!
• At this stage I duplicate the image to produce a Clone.
• The original I stretch with Histogram Transformation as far as I reasonably can; I find this results in a pastel coloured image but nice stars.
• The clone I stretch with Arcsinh as far as I reasonable can; I find this results in an overly colourful DSO with stars that are over-coloured/saturated.
• I then transfer both as 16 bit TIF to Photoshop and use the original as background and the clone in a hide-all masked layer.
• I then brush the clone into the original using a soft 50% opacity brush being very selective about which parts of the clone I select.
• Once happy I flatten the image and use the Dodge/Burn/Sponge tool to selectively enhance small parts of the image.
• I use Levels to make sure the background is about where I want it, and sometimes Curves to make some minor adjustments to the colour balance.
• I then finish off using the free Nik Dfine2 noise reduction either selectively or on the whole image.

All of this is what I do when processing images from my osc/dslr.

When using the unmodified Canon dslr I also follow 'Making Every Photon Count' advice and only use iso800 and maybe change the exposure settings depending on the target; it also means I only need iso800 calibration frames as well.

Hope this of interest.