Video: Processing M27 in Pixinsight

Garbz (Chris) · #1 04-09-2013, 05:58 PM

Hey guys, When I processed my M27 photo the other night I thought I'd try something and recorded a video of the processing. Not sure what I was thinking at the time but maybe this will help me learn from my mistakes or learn what works etc.

The resulting M27 picture is here: http://www.flickr.com/photos/10090242@N03/9671462468/

Anyway I thought I'd share the process I went through to make it incase it helps someone. Video is hopefully embedded below.

http://www.youtube.com/watch?v=trR5g3riAk8

Look a few posts below for another video it produced much better results.

The steps I went through are thus:

Opened single sub and debeyered.
Stacked all the subs together.
Compared the stacked image to the original single sub. I did this because I recently found that when I stuff up the stack it will actually add noise so comparing the noise in both images is a sanity check that the stacking worked well.
Basic crop applied, and image mirrored.
DBE used to eliminate background colours.
Histogram transform was applied to brighten up the image. Here I entered the median of each channel in the midrange and then entered 0.92 as the RGB/K midrange. This normally gives me a good starting point.
Next change the RGB working space to all 1s (apparently this helps some of the tools like ACDNR)
I extract the lightness channel and then use LRGB Recombination to add saturation to the image and do some colour noise reduction. I find this works better for overall saturation than the curves tool.
Use Dynamic PSF to determine the point spread function for deconvolution.
Build a star mask that I will use for a few tools.
Then I try in vain to get a decent deconvolution working but throw my arms up in frustration and settle for another way of sharpening the image.
I create another mask for all bright objects.
I move to ATrouseWaveletTransform and use it to sharpen the image by controlling the bias.
SCNR is applied though not really needed as there wasn't much of a green tinge in this image.
Then I move to ACDNR to reduce lightness noise.
I prepared another mask that was blurry but focused only on the nebula itself.
Again I used ATrouseWaveletTransform to sharpen the image, this time with a different set of parameters to make the nebula "pop" a bit more.
Still not quite poppy enough I also used LocalHistogramEqualisation to add a very subtle HDR effect. I'm not a fan of overuse of the effect so by the time I'm finished with the settings the result is quite subtle.
Finally a bit of playing with curves to get the background down in brightness and then I add a bit of saturation to the blue stars.
A quick rescale and I posted it on the web.

I wonder if anyone will find this helpful.

RickS (Rick) · #2 04-09-2013, 09:24 PM

Chris,

I'm sure some folks will find your notes helpful. Coming up with a basic workflow is the initial challenge with PI.

I'm not surprised you had problems with deconvolution. You need to apply it to a linear image, i.e. before you stretch it.

Cheers,
Rick.

LewisM · #3 04-09-2013, 10:01 PM

I'd really love that video at normal speed so I could follow what you are doing!!! I am slowly learning PI when time permits, and your results speak for themselves.

REALLY appreciate a normal rate (apart from the stacking), showing the checkboxes, values etc.

If you don't want to post it on YT, perhaps email it to my gmail or Dropbox?

Garbz (Chris) · #4 04-09-2013, 11:23 PM

Quote:

Originally Posted by RickS

Chris,

I'm sure some folks will find your notes helpful. Coming up with a basic workflow is the initial challenge with PI.

I'm not surprised you had problems with deconvolution. You need to apply it to a linear image, i.e. before you stretch it.

Cheers,
Rick.

OH... MY... GOD!!!

Just tried it again, and it worked beautifully. I got completely confused in the manual where it talked about applying it to a linear image at the same time while talking about a linear workspace. I thought the only critical factor was to set the Gamma 1.0 and give equal weights to R, G and B components. Turns out they were talking about both a linear Gamma in the Lightness as well as a linear image response.

I really should have been paying more attention when looking over your shoulder at Astrofest. I blame the beer.

Thankyou very much! Now I need to go and reprocess every photo I ever took.

Quote:

Originally Posted by LewisM

I'd really love that video at normal speed so I could follow what you are doing!!! I am slowly learning PI when time permits, and your results speak for themselves.

REALLY appreciate a normal rate (apart from the stacking), showing the checkboxes, values etc.

If you don't want to post it on YT, perhaps email it to my gmail or Dropbox?

http://www2.garbz.com/M27-processing-video.mp4 Please be kind to my ADSL modem, this may take a while for you to download. Approx 40MB.

Also I didn't remove the stacking but you can fast forward through that. In light of what Rick said above, don't pay too close attention to my PixInsight use. I clearly don't know what I'm doing.

Also apparently I only recorded it at 15fps so this is still faster than realtime but should give you a far greater idea of what's going on.

LewisM · #5 05-09-2013, 05:28 AM

Thanks

Garbz (Chris) · #6 06-09-2013, 02:33 PM

Ok REDO:

This video starts after the stacking process. Has steps in a slightly different order and produces a much better result:

Click here to load the video

Steps:

Crop
Dynamic Background Extraction (twice)
Linearise the RGB Working Space
Starmask (this took me a while the tool behaves differently on a linear image)
DynamicPSF to get my PSF for Deconvolution.
Deconvolution (this again took a while to get ringing under control and I had problem with dynamic range as well)
Non-linear histogram transform applied to bring the image to the correct brightness.
Extract lightness channel
Now here's something different. I worked on reducing noise only on the lightness channel use ATrousWaveletTransform. In this case I had to make a mask for a mask and applied it to the Lightness Channel before LRGBCombination. That way when I do LRGBCombination the lightness is noise free and the colour channels actually go through noise reduction in the process. There was a little to-ing and fro-ing and at some point I hit the undo button about 6 times and started again when I realised I had too much noise in light areas because my mask sucked. I restarted this step almost from scratch at around 22min in.
Did some Local Histogram Equalisation as well to make the thing pop a bit.
Finally a bit of colour saturation in the blue channel on the stars.
After I finished I went back and bumped the saturation up slightly.
Also not shown in the video I opened the image YET AGAIN and used the MorphologicalTransform tool to apply an erosion filter to the image. This tightened the stars in and thus made the nebula more pronounced.

h0ughy (David) · #7 06-09-2013, 05:30 PM

Chris - cant see this last video - its tagged private

Garbz (Chris) · #8 06-09-2013, 08:14 PM

Thanks Dave, Fixed it.