IKO - M33 LRGBHa - Processing Competition

[Grant]

We've just released the latest set of data from the Ikarus Observatory project here.

Following the wonderful entries from the first and second processing competitions we are running another to see who can get the most from this data and create the 'best' image! The winner will receive £150 of First Light Optics gift vouchers and two runners up will receive £50 of gift vouchers.

Details below:

• Closing date: 29th March 2021 @ 22:00
• Please post your entries into this thread
• Please only use the data we've released - don't blend it with other data or add to it with other sources
• Use what ever processing software and techniques you like, potential bonus points for sharing your workflow and techniques with others - that way, we all get to learn as well 
• Multiple entries are allowed but, please refrain form posting multiple attempts that are very similar - better to update an earlier attempt if it's just an incremental improvement but, it would be interesting to see totally different attempts using different narrowband blends etc...
• If you win or are a runner up, we would really like to use your creation in future marketing materials, on our website etc.. so by entering the competition, you are giving us permission to use your image in this way.
• Judging will be done by a small team of judges from the FLO team - It's completely subjective  but will be based on what image(s) we think do the best job of extracting the most from the data released and, look the 'prettiest' 
• We will announce the winners by the 4th April 2021 live on StarGaZine then on SGL

Thanks all and look forward to seeing your entries!

[alexbb]

I saw the tremendous amount of data that you shared so I downloaded it and gave it a try. Thank you for making it available.

Here's what I did (what I remember):

In PixInsight

• Luminance
  1. Deconvolved the luminance layer
  2. Applied some TGV denoise with an inverted duplicated layer, stretched more than STF
  3. Stretched it partially with arcsin stretch
  4. Stretched it more with Histogram Transformation
  5. Duplicated it and applied HDR with an inverted duplicated layer, stretched more than STF. Kept this layer mask
  6. Combined 5 and 6 at around 60-40
  7. Duplicated 6 and applied MLT for a star mask
  8. Applied Morphological Transform (MT) on 7 with dilation and erosion for 2 masks, subtracted the smaller stars from larger for a contour star mask
  9. Eroded the stars contour a bit, the larger ones more
  10. Removed the stars from the mask at 5 and applied convolution for a new mask.
  11. Applied LHE in more steps (256, 128, 64, 32 at 0.25 amount) with the mask at 10.
  12. Applied MLT mostly enhancing the larger wavelets (10-12 layers)
  13. Combined 6, 11 and 12 at a ratio that I liked
• RGB
  1. Combined RGB
  2. DBE
  3. Arcsin stretch
  4. HT stretch
  5. Duplicated 4 and HDR
  6. Combined 4 and 5
  7. Increased saturation in the highlights
  8. Maybe some ACDNR ?!
• Ha
  1. DBE
  2. Arcsin stretch
  3. HT stretch

Then moved to GIMP

1. Combined L as luminance (~60-70%) into RGB at a ratio that I liked
2. Increased saturation for the highlights and lowered for the background
3. Brighter stars have a magenta cast at the top-left and green cast at bottom-right

Back to PixInsight

1. SCNR and save
2. Invert, SCNR, invert and save (this kills magenta)

Back to GIMP

1. Imported the LRGB, the SCNR-LRGB as color layer and the Invert-SCNR-invert-LRGB as color layer
2. Added a star mask as a layer mask for the SCNR-LRGB and Invert-SCNR-invert-LRGB
3. Blurred a little the layer masks and shifted them towards top-left and bottom-right in small steps until the magenta and green casts disappeared from the stars
4. Merged the results, saved as LRGB_p01
5. Duplicated 4., added a grayscale mask, stretched heavily the mask and setting the black point high so the background became black in the mask. Blurred a little the mask. Set the black point high in the layer itself too so that the dark areas in the highlights became darker for an increased contrast.
6. Adjusted the saturation again
7. Saved as LRGB_p02. This was the final LRGB
8. Imported the Ha layer as lighten only and made it red.
9. Stretched 8 heavily and moved the black point high
10. Tamed down the ratio of the Ha layer, then merged over the LRGB
11. Crop and saved as M33-IKI-alexbb-LhRGB_p03_crop.jpg

Upload

Edit:

1. Moved higher the black point and upload
2. Crop and upload again

Edited March 5, 2021 by alexbb

[JSeaman]

 

I had another go with the new HA data (found it much easier to work with, thanks!)

 

 

Edited March 12, 2021 by JSeaman

[billhinge]

Two methods

1  - RGB stars  blended with HaLRGB galaxy with colour masked curves tweaks (looks darker here than  PI  despite SRGB profile?)

 

 

 

2- Images in full HaLRGB - normal & lighter versions

======================

• PI Dynamic background extraction on Ha
• Create new L  with old L plus Ha
• Create HaLRGB with SHO-AIP blend
• ArcSinhStretch
• Dynamic crop of uneven right and bottom edge
• Slight amount of LAB curves
• DarkStructure Enhance
• Sharpen
• Adjust  colour balances, AI Noise reduce & Super res in  Pixelmator

 

 

 

 

Edited March 26, 2021 by billhinge

[bottletopburly]

First attempt using Startools 1.7 

 Created synthetic luminance from L+R+G+B ,  created red channel Ha +R using equal exposure times , then loaded files together L, RGB  using compose mode . 
@jager945

Edit second image is using Startools 1.7 Thanks to Ivo @jager945and @happy-katLRGB processed and saved as tiff , then loaded HAv2 file auto dev bin ,crop,wipe auto Dev with roi and saved >load Ha file as red ,rgbRgb legacy,interpolation off ,save file ,load saved, Open Ha file non linear option >Layer mode >Lighten ,open saved lrgb file in foreground  /composite now shows Ha accents added to lrgb .

 

Edited March 13, 2021 by bottletopburly
Ha added to image

[matt_baker]

First attempt at this data. Again, really clean so thank you FLO and IKI.

I did have a slight issue with the Ha in that I couldn't get rid of a gradient that's faintly present in the image.

Things to work on next is trying to not destroy the star colour but I'm happy with the rest

Good luck everyone!

 

[Anne S]

Here's my version. Processed in Pixinsight.

DBE applied to Ha image. Pixelmath used to reduce stars and the nebulosity around the galaxy and to leave the small ha regions around the galaxy.

RGB masters combined then colour calibrated and DBE

Split channels and combined the red channel with the modified Ha image before recombining and making the final image nonlinear. Colour saturation enhanced.

Luminance had DBE applied and then converted to nonlinear. HDR applied to bring out the detail in the galaxy, then combined with the HaRGB image using LRGB combination.

Slight noise reduction and then stars slightly reduced to empasise the galaxy. SNCR to remove a slight green tinge. Final application of curves to increase contrast.

(If I remember rightly!)

Lovely data.

 

 

[Dean Hale]

1st time attempting anything like this but enjoyable for a lazy Sunday afternoon. 

Processed in Affinity photo.

How did i do it? Basically created adjustment layers with abandon and wanged filters here there and everywhere then started again. Then used this tutorial and some help from the Affinity photo book by Dave Eagle. Looking forward to seeing the other submissions so i can improve.

Loaded the .tifs into individual layers
Used recolour
Levels and curves readjustment x2
Background filter
Denoise
Selective colour
Brightness x2
Clarity

For attempt 2 - The Christmas card look. Same process as above except also used Autostretch1(RGB) from the Photokemi Tools plugin and lots of selective colour adjustments within Affinity photo. 

Edited March 9, 2021 by Dean Hale
2nd photo

[dannybgoode]

Blimey - I hope I have done this lovely data justice.  Still pretty new to processing and PI in particular however it is very nice to be able to work with quality source material.  Trying to balance detail without oversaturating the colours or destroying the fine detail in the dust lanes was a real challenge.  Loosely my workflow was as follows:

PixInsight;

EZDenoise and EZDecon the LRGB data.  I also ran DBE on the B however it did not seem to have a positive effect of the L R & G so I didn't repeat on these.

LRGB combination playing with the various intensities to get a decent base image.

DBE and EZDenoise on the Ha and then used the NBRGB script to add in the Ha data.

EZHDR and EZ Star Reduction and the much playing with the histogram levels for each R G B individually.

EZ Soft Stretch and then more playing with the histogram levels for each R G B again.  So much subtle detail in the dust lanes and getting this balanced with having a bit of punch to the blue and red was hard for me.  As was the red tinge to the background as a result of adding in the Ha data.  I wanted to minimise this without upsetting the colour balance too much.

Overall it took me 3 or 4 attempts in PI to get a base TIFF I was happy with to shove into other software for final editing.

DxO PhotoLabs 3;

PL3 has a brilliant colour wheel feature that seems to work differently to the HSL sliders in LR.  It makes tweaking saturation and luminosity a doddle.  So totally removed the green and enhanced the blues and red.

Added just a hint of what DxO call ClearView.  It's a bit of an HDR type effect and is very effective for daylight terrestrial photography however at default settings it is too much for astro work so just added in a hint.

Little bit of a play with vibrancy and saturation also.

Finally I added in a number of control points around the main structure of the galaxy and used this to null out the red in the background whilst leaving the main nebulosity of the galaxy intact.

PS CC;

Finally into PS and applied a few Astronomy Tools actions - Enhance DSO, Deep Space Noise Reduction and Make Stars Smaller

Then a tweak of contract and vibrancy & saturation and I think that I am done.

[matt_baker]

Second full attempt at the data which I've just finished.

May edit when I do incremental upgrades

Matt

 

[Spongey]

First attempt with this incredible data... It's great having some quality broadband stuff to work with  

Processed in PixInsight and Lightroom.

Just a smidge of Ha was added using this method; the LRGB data alone looked great in the Ha regions so I didn't want to push it.

Edited March 7, 2021 by Spongey

[carastro]

Right out of my comfort zone with this one.   Get much more practice processing narrowband due to location.   Great exercise though, I finally learnt how to add Ha to galaxies and make it show. 

Processed in Photoshop.  Mostly levels and Curves and selective colour.  Gredient exterminator to the Ha. 

Edited March 16, 2021 by carastro
2nd try

[dannybgoode]

EDITED - new Ha data used and did a little work in DxO.  Workflow updated.

Had to give this another go.  Every day I seem to learn something new about PixInsight and I wanted to do more of the heavy lifting there.  I am acutely aware I have been very reliant on Photolabs, LR and PS to fix flaws with my final image from PI instead of subtly enhancing it and I wanted to try and change this.

So, armed with some new knowledge I had put myself to learning overnight here is attempt two with the following workflow:

PixInsight;

For each R G B I ran ABE and for reasons unknown it was better behaved today than yesterday.  Then EZ Denoise and I lowered the settings a bit for it to be less aggressive trading a touch more noise in return for retaining a little finer detail.  5000 iterations though so each channel took some time.

For the L I ran ABE and EZ Decon but left out noise reduction this time - again trading NR for more detail which I felt important in the scheme of things, particularly with the L data.

Then for each LRGB I ran Masked Stretch and then used the RGB combine process to have the initial image.  Now for a new trick to me as I had very much neglected masks in PI however using the lum data I blew the galaxy core to almost all white using Histogram Transformation and then applied this as a mask.  Using ColorSaturation I pumped up the colours in the galaxy and stars and was pleased to see how well the mask worked and left the background pretty much untouched.

Ha followed the same as R G B however I applied a smidge of convolution just to soften the edges a bit with a view to more subtle blending when adding the data in using the NBRGB script.  I felt I overblew the reds far too much first time round so decided to try a less is more approach.  I used Starnet++ to create an Ha mask and applied this to my RGB image and then ran the NRGB script.  The Ha data is strong in this one so scale was right down at 0.4.

To add the Lum I ran PixelMath to create a 15% gray mask which I applied to the HaRGB image and used Channel Combine to add in the Lum.  The mask just tones down the Lum a bit to avoid blowing the highlights too much

Then Photometric Colour Calibration and played with each R G B in the Histogram Transformation process just to balance the colours a touch.

SCNR next - full whack to get rid of the green tinge and then just a smidge on the blue (Amount = 0.25).  I was battling the colour tinge far far less this time round anyway and that just did the trick.

A couple of final steps, first a hint of CurvesTransformation just to increase the contrast a smidge and then reapplied the saturation mask and tweaked it until I was happy.

I them merged my second attempt with this one using PS CC.  I also added just a hint of some diffraction spikes to some of the stars.  I have the PI only image as a bottom layer and then added the v1 image and applied the diffractions using Astronomy Tools and dropped the opacity down to 60% but boosted the saturation.  I then dropped another layer of the PI image on top and dropped the opacity on that down to 65%.  Overall I think this about balances the Ha and the stunning sparkling blues of the galaxy whilst retaining a soft overall look with plenty of detail.

Finally into DxO as I wanted a little punch whilst retaining the dust lanes etc as far out into the arms as possible.  DxO has a feature called ClearView plus which is incredibly effective when dealing with terrestrial images.  It is not entirely unlike HDR but adds punch in a different way.  However, it is usually too much at default settings for DSO's etc.  So I added in just a hint and then used the colour wheel feature to selectively add a bit of saturation and to change the luminance of the blues and reds.  The colour wheel works quite differently to LR's HSL slides and can remove or enhance colours as you see them and does not subtract them from the overall RGB mix.  Very handy for killing visible greens without changing the rest of the colour balance..  A touch of vibrancy and saturation, a tweak of NR and contrast finally some control points on the outer rim of the nebulosity to control some colour tinges.   I am now calling this done.

Honestly there is so much in this data and it's fascinating to see so many different interpretations of it however I fear madness descending if I continue to faff and fiddle.

Overall this left me with a final image that I was pretty pleased with - certainly my best attempt at getting a decent image just out of PixInsight and then minor tweaks in PS and DxO.   Certainly not something I would have dreamt of being able to do 48 hours ago...

 

 

 

 

 

 

Edited March 9, 2021 by dannybgoode

[ollypenrice]

Two renditions, the first with a minimal application of the Ha data, the second with it used selectively but more emphatically. Lovely data, beautifully calibrated. Many thanks.

 

 

 

RGB combined at parity in AstroArt 5 and run through DBE in Pixinsight. This was hardly necessary because the data were so clean and gradient-free but it does set the background to neutrality in the three channels. Very light application of SCNR green. (I had one small patch of blue noise which I fixed with SCNR set to Blue - a first for me! I saved it and applied it as a bottom layer in Photoshop, taking the problem patch off with the eraser.) RGB given a simple log stretch in Photoshop CS3 (Levels.)

Luminance given DBE in Pixinsight (again it hardly needed it) then given an initial log stretch in Photoshop CS3 till the background was at 21. After this I pinned the background at that level and continued to stretch just above that in Curves. Initially I did a very hard stretch (which the data supported) but in the back of my mind I knew it would overwhelm the RGB - which it did! So I redid the Curve in Luminance, stopping short of what it would support, but kept the hard stretch for later.

L applied to RGB in Photoshop and then the LRGB was further stretched as one in Curves with the background pinned to 21 again. I then applied my over-hard stretch as a luminance layer at low opacity to get what I could out of it.

Next, lots of tiny iterations to tweak colour, reduce stars (Noel's Actions) and add contrast by pinning points in Curves and stretching just above that. I went back into PI to run LHE at one point but brought it back into CS3 to use as a Layer at partial opacity.

Core sharpened as a bottom layer using Unsharp Mask with the stars excluded. Improved bits were let through using the eraser on the top layer.

A few large stars were reduced using Ps Layers: Make copy layer and use circular well feathered eraser to take off top layer over and just around the star in question. Activate bottom layer, go to Curves and pin the curve at the background level just around the star. Then pull down the curve just above that. Reducing the star increases its saturation so reduce that. I cannot recommend this technique on big stars too highly.

Noise reduction: I was tempted to use none at all because the data were ultra-clean and M33 has a strangely grainy, noise-like speckled look even in the Hubble image. However, I ran Ps Despeckle as a bottom layer and took off the top layer only for the faint glow between the spiral arms. The background sky and brighter parts had no NR whatever. Well done IKI!

Ha had a gradient! (Moonlight for a guess?) That was removed in PI/DBE. Then stretched brutally using an aggressive Curves stretch in Ps. (Not trying to make a nice Ha image, not worried about the noise in the darker parts because they won't end up in the final blend. I just wanted to get those nice ring structures above the brightness of the red channel so I could let them into the final image.)

Ha then added to red in blend Mode Lighten.  The resulting image was placed as a bottom layer below the LRGB and allowed into the image where desired. (I was looking for small, delicate features absent from the LRGB. I didn't want a big red glow in the core.)

I think the Ha makes an entirely different image, neither better nor worse, so I've posted both.

Great fun and thanks to all involved for the data.

Olly

 

 

 

Edited March 11, 2021 by ollypenrice
Incessant tinkering!

[Neko]

This one is as challenging as fun!!!  

This is my first go, perhaps went too far with sharpening, but will try once again! 

I thought it was a gradient too, but then I've realised this is around 100hour integration time..and it could be nebulosity? So I've processed it this way. 

StarTools 1.7, Compose module. Merged Halpha + Red channel, and wiped cropped all files before loading them in the module, then sort of normal workflow

finished in LightRoom. 

 

[Grant]

The Ha is a bit wonky - I've redone it by itself and been more aggressive with ditching some frames - I've uploaded it to http://ikiobservatory.com/data/M33_2020-03-05/ as the 'v2' ha files.

Hope that works better - sorry about that!

[emyliano2000]

This is my attempt with the original Ha file.

I've done so many things that I couldn't keep count.

I combined the LRGB in Astropixel processor, calibrated the star colour, Moved to pix for DBE, soft stretch, saturation and a few other things. Moved to photoshop where I blended the Ha, used the luminance again, Added contrast to selected parts, high pass filter, vibrance, noise reduction and so on.

Emil

 

Edited March 16, 2021 by emyliano2000

[Padraic M]

Lovely data to work with. Processed in AstroPixelProcessor to remove gradients/LP, and combine as LHaRGB.

Selective masking and curves in Gimp, plus a little contrast enhancement in the central spiral arms.

Finally into Noise Ninja for a little noise reduction - very little needed really, and that sandy texture in the centre has to be protected.

 

[Neko]

Second Try!  Almost same process as before

StarTools 1.7 and Lightroom. 

[Spongey]

Thanks for the updated Ha stack @Grant, it is much cleaner and easier to process.

Attempt number 2 with the new Ha stack follows. I tried to bring out a bit more of the Ha regions in this process, both in luminance and colour.

Cheers

 

Edited March 9, 2021 by Spongey

[almcl]

Using the V2 Ha data and processed in StarTools 1.7.485, layering the Ha data into the LRGB image after separate processing.

[tomato]

Here is my effort, using the V2 Ha data. Processed in StarTools V1.7 first making a LRGB composite, then blending in the Ha using Ivo's tutorial, except I used the Distance option in the Layers module, rather than Lighten. Final adjustments in APP, mainly using the HSL tool to try and accentuate the dust lane details.

The data is, as with previous sets, absolutely first class, I was hoping LRGB would be more straightforward than NB, but there is so much depth and detail, I don't know where to start and when to stop. The Ha data is amazing, so much going on at this wavelength in M33.

 

 

[TerryMcK]

Here is mine processed in PixInsight

• Combined rgb with ChannelCombination in rgb mode
• Combined lum to rgb with ChannelCombination in HSI mode
• Ran Dynamic Background extraction
• Ran Photometric Colour Calibration
• Histogram transform after tweaking the STF slightly
• Ran Starnet on this image to remove stars
• Created a rangemask and applied it to destarred image
• Inverted the mask to project the galaxy core and then tweaked the saturation, luminescence slightly to bring out some colour
• Next stretched the Ha image and removed the stars
• Ran the script NBRGB combination to add the Ha to the red channel
• Denoised the NBRGB_Combination image
• Ran the DarkStructure Enhance script
• Ran a local histogram equalization twice on the image with slightly different parameters
• Used PixelMath to add the RGB StarMask back to the NBRGB Image
• Called it finished

 

Edited March 13, 2021 by TerryMcK

[gorann]

On 08/03/2021 at 17:02, ollypenrice said:

Two renditions, the first with a minimal application of the Ha data, the second with it used selectively but more emphatically. Lovely data, beautifully calibrated. Many thanks.

RGB combined at parity in AstroArt 5 and run through DBE in Pixinsight. This was hardly necessary because the data were so clean and gradient-free but it does set the background to neutrality in the three channels. Very light application of SCNR green. (I had one small patch of blue noise which I fixed with SCNR set to Blue - a first for me! I saved it and applied it as a bottom layer in Photoshop, taking the problem patch off with the eraser.) RGB given a simple log stretch in Photoshop CS3 (Levels.)

Luminance given DBE in Pixinsight (again it hardly needed it) then given an initial log stretch in Photoshop CS3 till the background was at 21. After this I pinned the background at that level and continued to stretch just above that in Curves. Initially I did a very hard stretch (which the data supported) but in the back of my mind I knew it would overwhelm the RGB - which it did! So I redid the Curve in Luminance, stopping short of what it would support, but kept the hard stretch for later.

L applied to RGB in Photoshop and then the LRGB was further stretched as one in Curves with the background pinned to 21 again. I then applied my over-hard stretch as a luminance layer at low opacity to get what I could out of it.

Next, lots of tiny iterations to tweak colour, reduce stars (Noel's Actions) and add contrast by pinning points in Curves and stretching just above that. I went back into PI to run LHE at one point but brought it back into CS3 to use as a Layer at partial opacity.

Core sharpened as a bottom layer using Unsharp Mask with the stars excluded. Improved bits were let through using the eraser on the top layer.

A few large stars were reduced using Ps Layers: Make copy layer and use circular well feathered eraser to take off top layer over and just around the star in question. Activate bottom layer, go to Curves and pin the curve at the background level just around the star. Then pull down the curve just above that. Reducing the star increases its saturation so reduce that. I cannot recommend this technique on big stars too highly.

Noise reduction: I was tempted to use none at all because the data were ultra-clean and M33 has a strangely grainy, noise-like speckled look even in the Hubble image. However, I ran Ps Despeckle as a bottom layer and took off the top layer only for the faint glow between the spiral arms. The background sky and brighter parts had no NR whatever. Well done IKI!

Ha had a gradient! (Moonlight for a guess?) That was removed in PI/DBE. Then stretched brutally using an aggressive Curves stretch in Ps. (Not trying to make a nice Ha image, not worried about the noise in the darker parts because they won't end up in the final blend. I just wanted to get those nice ring structures above the brightness of the red channel so I could let them into the final image.)

Ha then added to red in blend Mode Lighten.  The resulting image was placed as a bottom layer below the LRGB and allowed into the image where desired. (I was looking for small, delicate features absent from the LRGB. I didn't want a big red glow in the core.)

I think the Ha makes an entirely different image, neither better nor worse, so I've posted both.

Great fun and thanks to all involved for the data.

Olly

 

 

 

Both look great Olly and thanks for giving the detailed work flow. Most of it similar to what I would have done but one thing was new to me: I never tried PS Despeckle. I tried it now on a different image and got the feeling it was maily a light Gaussian blur, but maybe not. What is the advantage of it in your view compared to other NRs?

[ollypenrice]

25 minutes ago, gorann said:

Both look great Olly and thanks for giving the detailed work flow. Most of it similar to what I would have done but one thing was new to me: I never tried PS Despeckle. I tried it now on a different image and got the feeling it was maily a light Gaussian blur, but maybe not. What is the advantage of it in your view compared to other NRs?

I don't think there's anything special about it, Goran. Sometimes the noise texture just strikes me as likely to be well-handled by it so I try it. Essentially I tried it on a whim and it seemed about right and very mild. There was next to no noise in this data anyway, if you were careful not to provoke it during the stretches. The downsampling (I assume) of the data for publication left it superbly clean. I feel the trick is to stretch the backgound only to the desired brightness and then no more. Stretch above that, of course, for all it will give.

Olly