IC 1805 Heart Nebula in Cassiopeia
November 7, 2023
Green Lane Park, PA
Bortle ~5.4
© 2023 ubergarm

~2 hours 45 minutes total exposure
28 x 180s Hα
26 x 180s Oiii
flats darks and bias

AT80ED w/ AT80EDRF 0.8x
OGMA AP26MC @ 100g, Low Noise Mode, HCG, Normal Dynamic Range, BP=368, 0 degC, DirectShow Driver
ZWO AM5
Manual Focus
ZWO Filter Drawer
Antlia 4.5nm EDGE 2" Mounted Filters

WO UniGuide 50mm Guide Scope
ZWO ASI585MC "Guide Cam"

SharpCap 4.1 Beta LiveStack Capture (individual subs saved and post-processed)
PHD2 Guiding
Siril 1.2.0
GraXpert v2.0.2 AI Model v1.0.1
StarNet++ v2.0
~1.75 arcsec/pixel
~3°x2°

This is my first attempt at narrowband imaging and guiding. I've never imaged more than ~30 second subs before, but with 4.5nm filters longer subs are possible in my "dark sky" site. I had some hiccups dropping frames with this version of SharpCap, but fortunately Dr. Robin Glover is adding native driver support for the OGMA camera (Touptek OEM) coming soon:

https://forums.sharpcap.co.uk/viewtopic.php?p=39222#p39233

I'm not sure if it was better sky conditions, or if this object just has more Ha signal than nearby Soul and Lobster Claw / Bubble Nebula areas, but this image turned out "better" with less total integration than those other targets which I imaged more recently.

## Current Workflow

1. Organize all the calibration and light files into folders for my Siril script.
2. Generate individual monochrome stacks (weighted by wfwhm) for each filter e.g. Ha and Oiii in this case.
3. Align and re-normalize the Oiii to Ha then crop each one exactly the same to remove border artifacts.
4. Individually perform background gradient removal with GraXpert AI.
5. Individually perform Dynamic PSF and Deconvolution followed with a touch of De-Noise.
6. Use PixelMath to create a single RGB image from the separate monochrome frames e.g. Red = Ha, Green = 0.6 * Ha + 0.4 * Oiii, and Blue = Oiii
7. Color calibration background neutralization and white reference based on selections from the image.
8. Remove green noise (optional)
9. StarNet++ star removal using "pre-stretch linear image"
10. Many GHS (Generalized Hyperbolic Stretch) using Human-weighted or Even-weighted luminance to preserve the colors better.
11. Trim the black point as needed and stretch a little more
12. Do another color calibration background neutralization maybe
13. Saturation Stretch a little for more color intensity without going overboard
14. Recombine the starless image with the starmask with just a little modified arcsine stretch using human/even weighted luminance to preserve star color
15. Save to JPG and optionally add captions using a python script then upload.

## Future Ideas
As part of Step 2 Stacking:
I'd like to try PI (PixInsight) and APP (Astro Pixel Processor) or write python scripts using DrizzlePac for real Drizzle algorithm with adjustable pixfrac and kernel parameters. Siril's "-drizzle" feature is essentially pixfrac=1 square kernel upsampling. This algorithm was developed for use on the HST because it was imaging with pixel sizes roughly the same as the instrument's Dawes limit. This is the same situation for amateurs using wide field scopes in good seeing. With proper dithering it can improve spatial resolution at the cost of some noise (or just more subs). Perfect for smaller sized objects!

As part of Step 5 Deconvolution:
I haven't seen anything better at deconvolution than RC Astro's BlurXTerminator PI plugin. Siril's algorithms seem limited to a single PSF across the entire image. I'm looking into a state-of-the-art python script called PIFF: PSFs In the Full Frame that can model your physical scope and seeing conditions to deconvolve a linear combination of PSFs in sky-coordinates specific to each part of your sensor. For example, it may be able to take into account coma and other aberrations that vary across the corners of the image. It looks pretty complex though and would require tuning parameters for each rig. No wonder people like the easy button AI filters hahah...

As part of Step 10 Stretching:
Some folks have suggested creating a synthetic Luminance layer around this point. The idea is this Lum layer can be processed optimizing for sharpness over noise, and the remaining RGB layers can be processed for low noise. I haven't tried it, and haven't seen empirical evidence showing the potential gains vs the hassle. Despite not adding any more signal, it may work well given the psychophysics of human visual perception.

Thanks for looking and clear skies!
-ubergarm