PixInsight Image Integration

[groberts]

I've been battling to learn pre-processing for over a week now but have at last Calibrated, Corrected, Weighted etc subs for each LRGBHa wavelength, as instructed in the Light Vortex (LV) tutorial + all saved together in a single folder.  Whilst the LV tutorial is generally excellent here and in other sources I'm unable to fathom out how to handle the subs in Image Integration - I get how to process x1 wavelength but not all five as in this case.  

Based on my experience with DSS, at this point I would expect to stack each wavelength separately i.e. load L subs into Image integration + Apply Global / run etc to produce a stacked L image, then move on to the next wavelength and do the same again.  However, various sources seem to suggest I load all the subs (LRGBHa) and run them all together.  Contrary to my intuition I have actually done this and I did prodcue a result but I'm  baffled as to how to get form here to five stacked LRGBHa images or as I suspect, should I process each wavelenght one by one?

Graham      

    

[MarkAR]

Process the alignment and stacking separately just as you have with the previous steps.

Make sure that you use the same "best" image for alignment for each stack. 

Once stacked, open up and discard rejection high and low, save each separate stack.

You are now ready to post process. Some processes can be done before channel combination like Dynamic Crop, DBE, noise reduction.

[groberts]

Thanks Mark, that makes perfect sense but was not what various tutorials seemed to say. 

After something of an uphill battle I feel near the finishing line (at least for this stage) and will try again tomorrow.

Graham    

[wimvb]

2 hours ago, MarkAR said:

Once stacked, open up and discard rejection high and low, save each separate stack

I find "rejection low" a good indicator for the dynamic crop process

[x6gas]

The LV tutorials are carefully put together and are generally excellent but there are some bits where my experimentation and / or other advice has led me to a different process.  For example, I am not convinced by the weighting expression used in subframe selection (which in my view - or at least for the images I've been processing lately - doesn't take enough account of eccentricity) so I use Noise Evaluation instead.  I've also abandoned local normalisation...

Interestingly I've also found sigma clipping to give the best results on stacks of 30+ images...

[wimvb]

56 minutes ago, x6gas said:

Interestingly I've also found sigma clipping to give the best results on stacks of 30+ images

+1

 

57 minutes ago, x6gas said:

doesn't take enough account of eccentricity

I usually put that into the weighting formula. 

[x6gas]

3 minutes ago, wimvb said:

+1

 

I usually put that into the weighting formula. 

Yes - to be clear it is in the expression but for me it doesn't have a strong enough influence on the calculated weighting...

[wimvb]

27 minutes ago, x6gas said:

it doesn't have a strong enough influence on the calculated weighting...

I vary the coefficients in Kayron's formula. 50 x ecc + 25 x fwhm + 25 x snrw, depending on the data. Just make each of the three terms vary from 0 to 1 as in the tutorials. Then vary the proportions to taste. I've never done a thorough analysis, though. Weighting should decrease the snr in the final image as compared to plain averaging, afaIk. 

[groberts]

Thanks everyone, I hope to have some stacks later today!

Out of interest - at the end of the integration / stacking process when there are the high/low/new images to assess (a) have these been saved anywhere, or is that always a manual choice and (b) have any of the underlying calibrated etc subs used for stacking been changed by the process - the point being if you're not happy with the result can you just change the appropriate setting ther and then e.g. weighting and immediately re-run the otherwise same set-up again?

Also - given the recommendation to use different rejection algorithms depending on the number of subs, can you do this for different filters if the number of subs are very different and still easily combine the respective stacks generated later?     

Graham  

Edited June 2, 2020 by groberts

[wimvb]

2 hours ago, groberts said:

(a) have these been saved anywhere,

No, you have to save them. Normally you wouldn't save the rejection images.

2 hours ago, groberts said:

(b) have any of the underlying calibrated etc subs used for stacking been changed by the process

Depends, if you choose to over-write the originals and actively remove any pre- or post-fix, then the files are being over written. Otherwise each process creates its own files. After a successful integration process, you would normally delete intermediate files. Maybe keep the calibrated subs. Definitely keep your raw data.

2 hours ago, groberts said:

the point being if you're not happy with the result can you just change the appropriate setting ther and then e.g. weighting and immediately re-run the otherwise same set-up again?

Better still, if it's just the integration process you want to redo, pixinsight keeps the latest files in cache memory, and any subsequent restacking (integration) will be much faster.

2 hours ago, groberts said:

Also - given the recommendation to use different rejection algorithms depending on the number of subs, can you do this for different filters if the number of subs are very different and still easily combine the respective stacks generated later?

All the integrations are independent from each other. You can mix any rejection algorithm. Eg linear rejection for L, percentile for R, G, and B, sigma clipping for Ha.

[groberts]

Thanks Wim that's very helpful.  As you might discern, I'm influenced by years of working with DSS - this is a whole new ballgame. 

I've nearly (I hope) got x5 stacks now, so at last it's almost time to move on to post processing and probably more (a lot more) questions! 

FYI I'm working from Light Vortex + Warren Keller's book + new online version of Rogelio Bernal Andreo's new Mastering PixInsight, all of which together are excellent and provide extensive detail but there always seems to be something not quite clear and first hand + real time help is indispensable - thank goodness for SGL.

Graham

[x6gas]

If you fancy trying deconvolution I recommend Jon Rista's tutorial.  Never had any success with deconvolution using various software until I read this...

[groberts]

51 minutes ago, x6gas said:

If you fancy trying deconvolution I recommend Jon Rista's tutorial.  Never had any success with deconvolution using various software until I read this...

I'll certainly have a look at that when / if I emerge form PI but not sure I need any more processes at the moment, though Jon's work and insight on photography / astrophotgraphy in general is always worth consideration.

Graham 

[wimvb]

Deconvolution is a wonderful tool, but I would hold off on it until a later time. Once you have the data, you can process it over and over, as you learn pixinsight. For a novice, background extraction, background neutralization, colour calibration, stretching, noise reduction, and perhaps star reduction, are quite enough to wrap one's head around and avoid . 

Just my 0.02€

[x6gas]

30 minutes ago, wimvb said:

Deconvolution is a wonderful tool, but I would hold off on it until a later time. Once you have the data, you can process it over and over, as you learn pixinsight. For a novice, background extraction, background neutralization, colour calibration, stretching, noise reduction, and perhaps star reduction, are quite enough to wrap one's head around and avoid . 

Just my 0.02€

Well I found decon not too bad having read Jon's post - the first time I'd seen a systematic approach to the settings and it works for me - but you're right, it's another thing to wrap your head around.

I've had enormous trouble with ABE / DBE clipping the background sky for some reason.  Haven't figured it out yet so just not doing it!

[wimvb]

54 minutes ago, x6gas said:

Haven't figured it out yet so just not doing it!

Check normalization in the correction tab. This maintains the median background (as well as any colour cast, so you need to do background neutralization afterwards). 

[wimvb]

Here's an interesting reference, albeit a bit old

https://pixinsight.com/forum/index.php?threads/how-does-one-interpret-the-noise-evaluation-statistics-in-imageintegration.1933/

[x6gas]

4 hours ago, wimvb said:

Here's an interesting reference, albeit a bit old

https://pixinsight.com/forum/index.php?threads/how-does-one-interpret-the-noise-evaluation-statistics-in-imageintegration.1933/

Interesting read, thanks for sharing.