First image in PI: 32 hours narrowband on M1 [Reprocessed]

[astrovienna]

[Reprocessed:  http://www.pbase.com/skybox/image/159811685
see the post at the bottom for details]

First image in PixInsight.  I put this move off about two years too long, I think.  I'd appreciate any comments or suggestions.  This is a tricolor narrowband image of M1, with about 32 hours total time.  (I would have posted this in the gallery, but the rules there said nothing except 100% PI, and you'll see I used PS a tiny bit at the end here.)

http://www.pbase.com/skybox/image/159811685

If anyone wants to take a crack at the data, the stacked images for each channel are here:

https://dl.dropboxusercontent.com/u/58468743/SII%20from%20PI%20Reregistered.zip

You'll see the frames didn't completely align, because for some reason I can't explain I shifted orientation by 45 degrees midway through.  Duh.

Here's my procesing routine in PI.  (Thanks to Harry for the videos - a huge help up the learning curve!):

1.   Batch Preprocessing for calibration
2.   Star alignment
3.   Image Integration
4.   Channel Combination, straight Hubble/SHO.  I  usually do some color shifting at the end to tone down the green, and I experimented with SCNR at the end,  but I decided I kind of like the green in this image.  Let me know what you think of the color. 
5.   DBE
6.   Background neutralization (I saw almost no change)
7.   Histogram Transformation and Curves Transformation
8.   ACDNR.  I compared ACDNR, ATWT (applied before stretching), and TGV and really saw no difference not accounted for by different aggressiveness settings.  A bit more histo after this.
9.   Prepared nebula mask using ATWT. (much thanks to Rob77 from CN for the help.)
10.   Deconvolution, applying the inverted nebula mask so that only the nebula was sharpened.  I really liked the deconv tool.  It seems to have more precision than the AstraImage PS plug-in I've been using for years.  I have to test this out on my planetary images.
11.   Color calibration for star colors.  Masked out the nebula, selected almost all the stars in the image with previews and Preview Aggregator.  The stars were improved quite a bit, though perhaps still a bit too greenish.  But this also made the stars too light overall.  I'm not where I was going wrong.  So, to deal with that:
12.   Port into Photoshop, layering sharpened and unsharpened nebula images, then masked out embedded stars so they were not sharpened.  I couldn’t see an easy way to protect the embedded stars from sharpening in PI.  My PI star masks seemed too weak on embedded stars, so in PS I just dotted the mask over the stars to protect them.  Then I layered the image with the color calibrated stars and used it for star color, not star luminosity, to deal with the star dimming I was getting in PI.  I also fixed the registration errors in the corners with masked histo stretches and cloning.
13.   Save as TIFF, re-port into PI, slight histo adjustment in PI for black point.
14.   Downsize to 75% original.

I'm sure I did a zillion things wrong, but I'm happy with the result.  PI really is a great tool. 

Kevin

[RickS]

Quite a nice result, Kevin.  Thanks for the data.  I might have a play with it if I get time.  I'll let you know if I do...

Cheers,
Rick.

[oldwexi]

Hi Kevin!
You created a beautiful image!
Thanks for providing the original data.
Cant do it better...

Thanks that you described what you were doing in PS.
This brought me to the following lines:

The big adavantage of deconvolution in PI is that you can sharp nebula AND stars and also
reduce the elongation of the stars. There is no need to jump into PS - you can do the things
better in PI.

So, i did a simple try  with your data.
1.Channel Combine with PixelMath
2. Deconv with Starmask using in the Blend script the Starmask function
    (The Blend Script allows sharper better tuned starmasks!)
3. Step 2 again
4.) Masked Stretch
5.) Histogramm to neutralize Background
6.) Deconvolution(!!) to reduce elongation of all stars
7.) LHE to increase contrast in bright areas
8.) Curves to darken the Background
9.) Resample 75% to get same size..

Did not do any noise removal, sharpening or cosmetics so far
example here: http://www.werbeagentur.org/oldwexi/fotos/M1_rgb_gw.jpg
RGB-Version here: http://www.werbeagentur.org/oldwexi/fotos/M1_rgb1_gw.jpg

Basically with the first 6 steps you get what you want without
manual selective actions against the data. Deconvolution
does the sharpening (removal of seeing) from the beginning and  there is no need
using sharpening tools outside from PI.

Gerald
 

[astrovienna]

Thanks for the comments, Rick and Gerald.

Gerald, I definitely like your processing.  Can you describe Step 2 a bit more?  I'm not sure what the Blend script is.  And in step 7, what did you do to deal with star elongation?  First time with PI, so I'm not sure how deconvolution is really controlled.

Kevin

[Geoff]

And in step 7, what did you do to deal with star elongation?  First time with PI, so I'm not sure how deconvolution is really controlled.

Kevin

Get started version of decon:
http://mike-wiles.blogspot.com.au/2013/08/pixinsight-deconvolution.html

Upmarket version
http://pixinsight.com/examples/M81M82/index.html

[oldwexi]

Hi Kevin!

The Blend Script is a script which allows Blending Mode Functions in PI which you maybe know from PS.
Authors: Mike Reid, Hartmut Bornemann and the Starmask Mode by me.
So, i have added the possibility to create a starmask by selecting in the blending "Overlay Mode:" box "StarMask".
The basic idea of this starmask creation is to find with pixelmath the leading edges of the stars in the image. Based on the selected
threshold which defines the start of the slope.
Here the download link:
http://www.werbeagentur.org/oldwexi/Daten.Blend_StarMask_english.zip

Concerning Deconv in Step 6, i let all the Parameter at Std except:
set the "Aspect Ratio: 0.42   (whch reduces the horizontal elongation of the stars a little).
StDev: 0.5
Shape 0.7
Rotation 0.0

Hope this helps

Gerald

[RickS]

Here the download link:
http://www.werbeagentur.org/oldwexi/Daten.Blend_StarMask_english.zip

Hi Gerald,

I get a file not found using that link.

Cheers,
Rick.

[astrovienna]

Thanks Geoff and Gerald.  Geoff, thanks for the link to the great tutorial.  Very clear.  Gerald, I'll look into the script.  I'll also poke into the aspect ratio in deconvolution.  I completely missed that.

One more question:  is it better to do deconvolution separately on each channel, or after the color combine?  If I do it before, I assume I'll get different (and thus more accurate) PSFs for each channel, but will I then wind up with colored halos after the color merge?  Time to experiment!

Kevin

[oldwexi]

Upps!
Had a typo with link the correct link is:
http://www.werbeagentur.org/oldwexi/Daten/Blend_StarMask_english.zip

Gerald

[RickS]

Thanks Gerald. That's better!

[topboxman]

Here is my quick and dirty process.

A) RGB (SHO) combination.
B) Dynamic Crop to exclude any black/gray borders due to multiple nights framing
C) Save Dynamic Crop process icon
D) Get rid of RGB combined image
E) Apply the saved Dynamic Crop Process icon to individual subs and crop each sub.
F) Use Ha cropped sub as a reference for Linear FIT and apply Linear FIT to other two cropped subs
G) RGB (SHO) combination of cropped subs.
H) DBE
I) BN
J) CC
K) MMT to remove some noise while image is linear
L) HT
M) Create Star Mask for HDR
N) HDR using 6 layers.
O) LHE using Contrast Limit of 1.2
P) ACDNR
Q) Curves Transformation using "S" and "RGB/K".

No other sharpening was done. There is so much data that it was quite a pleasant to process this image.

Peter

[astrovienna]

Hello Peter!  I told you I'd get here one day!  Thanks to you and Gerald for processing the data.  I was wrong - it was actually 36 hours, not 32.  Fall seems to have that perfect combination of long nights and clear skies that lead to these big data sets.  In 2013 it was 59 hours on M76.  If only I could be so lucky in galaxy season.

The biggest thing I've learned from this is that there are endless ways to process in PI, and I should give up trying to find the "right" way.  Instead I'll just concentrate on finding "better" ways.

Kevin

[topboxman]

Hi Kevin,

Note the way I originally did Linear FIT and cropped the images was an overkill. I was worried that applying Linear FIT to your original images would not work well with the black/gray borders due to multiple nights imaging but I was wrong. I could have simply started by applying Linear FIT of your original subs and then RGB Combination and go from there. The first seven steps I wrote could have been reduced to three steps:

A) Linear FIT of original subs
B) RGB Combination
C) Dynamic Crop

Also, note that I didn't do any SCNR like I normally do for other narrow band images to remove excessive green from Ha applied to green channel. Crab Nebula has tons of data for Ha, Sii and Oiii that it was already colorful after RGB Combination.

I cropped the heck out of it so that DBE, BN and CC would not take into account of black/gray borders.

Maybe next time, try to get the same framing for multiple nights imaging. Did you have a hard time finding guide stars (due to using OAG at long focal length) which may have caused you to frame differently?

Peter

[astrovienna]

Hi Peter,

No, the bad framing was just a complete flub on my part.   But the data from that night was very good, so I wanted to use it.  SGP has a "manual rotator" feature that I finally got tuned into that helps get highly repeatable framing, and I'm trying to be good about using it.   

I confess I don't know what Linear FIT is.  Deconvolution is the only thing I've done yet outside of Harry's videos.  I too usually would do SCNR in narrowband to reduce green, but straight HSO seems to work well here. 

Kevin

[topboxman]

Linear FIT matches the brightness levels of images based on the selected reference image. I selected Ha as a reference image for Linear FIT and applied to other two images. In other words Linear FIT helps to equalize all three images. I normally use it for narrow band images because usually one NB image (typically Ha) would have much more information or data than other two (typically Oiii and Sii) so I equalize all three images using Ha as a reference as the first step of processing. I have seen people use Linear FIT for non-linear processing but that's another story.

I don't normally use Linear FIT for RGB. It always depends on the DSO and like you said, there are many ways to process with PI and sometimes no two DSOs can always be processed exactly the same way.

Peter