Rosette to Cone widefield mosaic

[MartinFransson]

Started this in 2017 and finally got some O3 for my mosaic of the Monoceros area.

Two panels of Ha and O3, combined to look more like an RGB image.

Ha: 27+27*5 minutes
O3: 29+31*5 minutes

Ha collected at f/2.8, O3 at f/4 for tighter stars. Samyang 135mm lens.

Thinking about collecting some LRGB but I find it hard to blend that in without the stars overpowering the image. Thoughts?

[vlaiv]

25 minutes ago, MartinFransson said:

Thinking about collecting some LRGB but I find it hard to blend that in without the stars overpowering the image. Thoughts?

I think that following approach is worth a try (you can even try it out with this data set to see how you like it):

Make starless version of both OIII and Ha with starnet++ and combine them to produce color version of nebulosity. From Ha stretched version and it's starless counterpart extract stars (subtract layers).

In this version just layer stars on top of combined nebulosity with lighten blend mode (this will give you pure white stars).

In RGB version collect only RGB data as you won't need luminance. Use stars only as luminance and do RGB ratio to transfer color to those stars. Again, transfer RGB colorful stars on top of above nebulosity.

In both bases (with RGB and without) your stars will be as tight as in Ha image, and in most cases those stars tend to be fairly tight.

[MartinFransson]

23 minutes ago, vlaiv said:

I think that following approach is worth a try (you can even try it out with this data set to see how you like it):

Make starless version of both OIII and Ha with starnet++ and combine them to produce color version of nebulosity. From Ha stretched version and it's starless counterpart extract stars (subtract layers).

In this version just layer stars on top of combined nebulosity with lighten blend mode (this will give you pure white stars).

In RGB version collect only RGB data as you won't need luminance. Use stars only as luminance and do RGB ratio to transfer color to those stars. Again, transfer RGB colorful stars on top of above nebulosity.

In both bases (with RGB and without) your stars will be as tight as in Ha image, and in most cases those stars tend to be fairly tight.

Thank you! I´m not sure I follow what you mean by "do RGB ratio to transfer color to those stars" but I´ll give it a try  I use Pixinsight and Photoshop.

[vlaiv]

3 minutes ago, MartinFransson said:

Thank you! I´m not sure I follow what you mean by "do RGB ratio to transfer color to those stars" but I´ll give it a try  I use Pixinsight and Photoshop.

It's fairly simple technique. I'll describe it in "pixel math" terms, but you can accomplish it in Photoshop with layers as well.

You take your R, G and B subs, wipe them (this is not needed if you aim only for star color, but if you do regular LRGB composing - it should be done) - meaning remove background and gradients from LP, and do color balance (however you do it - whether it is single star / multiple star color calibration or color ratio or whatever).

Next thing you do is add small offset to each of channels. It needs to be same and needs to be just enough so it moves all pixel values to be positive (this is in case that your wipe puts background at 0 and due to noise there are some negative values, you don't want any negative values in any of your subs, and also adding small offset reduces color saturation of background which is good thing as it reduces color noise in background sky if you do full LRGB image - again, not as important for stars only).

Then you stack those three channels with max function (not average but max). Next you divide each of colors with this max stack. This has prepared your color data for "transfer".

Next you take your stretched luminance layer and just simply multiply it with each of color ratio channels to give you proper R, G and B channel of final image. Btw - stretched luminance needs to be in 0-1 range, and above process of creating ratios will certainly put each ratio sub in 0-1 range.

Not above is not strictly correct way to do it in terms of color accuracy. Proper way to do it would be to first apply inverse gamma to lumiance, then multiply then apply proper gamma to result (multiplication needs to be done on linear data) - but that will create proper colors and most people will think that something is wrong with your image as they are used to much more colorful and saturated images . Most images in proper color look rather dull compared to renditions that we usually see and are used to it.

[MartinFransson]

28 minutes ago, vlaiv said:

It's fairly simple technique. I'll describe it in "pixel math" terms, but you can accomplish it in Photoshop with layers as well.

You take your R, G and B subs, wipe them (this is not needed if you aim only for star color, but if you do regular LRGB composing - it should be done) - meaning remove background and gradients from LP, and do color balance (however you do it - whether it is single star / multiple star color calibration or color ratio or whatever).

Next thing you do is add small offset to each of channels. It needs to be same and needs to be just enough so it moves all pixel values to be positive (this is in case that your wipe puts background at 0 and due to noise there are some negative values, you don't want any negative values in any of your subs, and also adding small offset reduces color saturation of background which is good thing as it reduces color noise in background sky if you do full LRGB image - again, not as important for stars only).

Then you stack those three channels with max function (not average but max). Next you divide each of colors with this max stack. This has prepared your color data for "transfer".

Next you take your stretched luminance layer and just simply multiply it with each of color ratio channels to give you proper R, G and B channel of final image. Btw - stretched luminance needs to be in 0-1 range, and above process of creating ratios will certainly put each ratio sub in 0-1 range.

Not above is not strictly correct way to do it in terms of color accuracy. Proper way to do it would be to first apply inverse gamma to lumiance, then multiply then apply proper gamma to result (multiplication needs to be done on linear data) - but that will create proper colors and most people will think that something is wrong with your image as they are used to much more colorful and saturated images . Most images in proper color look rather dull compared to renditions that we usually see and are used to it.

Do you mean that each individual sub in R, G and B has to be processed (wiping, offsetting etc) and then stacking them? Or is that a process for the stacked R, G and B images? Seems like A LOT of work to process each individual sub! 😮

Honestly, I think this process is above my skill level at the moment  

[vlaiv]

1 minute ago, MartinFransson said:

Do you mean that each individual sub in R, G and B has to be processed (wiping, offsetting etc) and then stacking them? Or is that a process for the stacked R, G and B images? Seems like A LOT of work to process each individual sub! 😮

Honestly, I think this process is above my skill level at the moment  

No, no, it is fairly simple, you start with stacked linear raw color images. Let's call those R_raw, G_raw, B_raw. Those are images that you would do regular RGB combine and stretch to create usual image.

You wipe them so that you remove any gradient and to remove any color cast due to sky flux. Then you color balance them (you don't have to, but if you do - this is the stage when it should be done if you want to do it properly).

This creates R_regular, G_regular, B_regular. Next you create ratio images as:

R_ratio = R_regular / max(R_regular, G_regular, B_regular)

G_ratio = G_regular / max(R_regular, G_regular, B_regular)

B_ratio = B_regular / max(R_regular, G_regular, B_regular)

simplest way to get above is to create mini stack of only 3 subs - three channel images and stack them with max function. This is only stacking involved and it is not regular stacking - it is just trick to get max values out of three images.

In the end ratio combination goes like this:

R_stretched = L_stretched * R_ratio

G_stretched = L_stretched * G_ratio

B_stretched = L_stretched * B_ratio

and that is it.

If you want to be correct in your color handling than you need to modify above with two changes.

1) when color balancing, you need to make sure you are color balancing to sRGB linear color space

2) Final combination should account for sRGB gamma in following way: R_stretched = gamma(inverse_gamma(L_stretched) * R_ratio) instead of R_stretched = L_stretched * R_ratio

That is it.

[MartinFransson]

6 minutes ago, vlaiv said:

No, no, it is fairly simple, you start with stacked linear raw color images. Let's call those R_raw, G_raw, B_raw. Those are images that you would do regular RGB combine and stretch to create usual image.

You wipe them so that you remove any gradient and to remove any color cast due to sky flux. Then you color balance them (you don't have to, but if you do - this is the stage when it should be done if you want to do it properly).

This creates R_regular, G_regular, B_regular. Next you create ratio images as:

R_ratio = R_regular / max(R_regular, G_regular, B_regular)

G_ratio = G_regular / max(R_regular, G_regular, B_regular)

B_ratio = B_regular / max(R_regular, G_regular, B_regular)

simplest way to get above is to create mini stack of only 3 subs - three channel images and stack them with max function. This is only stacking involved and it is not regular stacking - it is just trick to get max values out of three images.

In the end ratio combination goes like this:

R_stretched = L_stretched * R_ratio

G_stretched = L_stretched * G_ratio

B_stretched = L_stretched * B_ratio

and that is it.

If you want to be correct in your color handling than you need to modify above with two changes.

1) when color balancing, you need to make sure you are color balancing to sRGB linear color space

2) Final combination should account for sRGB gamma in following way: R_stretched = gamma(inverse_gamma(L_stretched) * R_ratio) instead of R_stretched = L_stretched * R_ratio

That is it.

OK, that sounds like I might be able to try  Just one thing, how do I colour  balance a single (mono) channel? In my world only an RGB image is possible to colour balance...

Thank you for taking your time to explain!

[vlaiv]

Just now, MartinFransson said:

OK, that sounds like I might be able to try  Just one thing, how do I colour  balance a single (mono) channel? In my world only an RGB image is possible to colour balance...

Thank you for taking your time to explain!

You color balance 3 mono frames - it's like color balancing regular color image but you do it on individual channels.

Simplest way to explain what is going on and explain how to do it would be:

take R_raw, G_raw, B_raw and do RGB compose to get Color_raw then color balance that into Color_balanced and then do RGB split to get R_balanced, G_balanced, B_balanced.

How it actually works is - with pixel math:

R_balanced = c1 * R_raw + c2 * G_raw + c3 * B_raw

G_balanced = c4 * R_raw + c5 * G_raw + c6 * B_raw

B_balanced = c7 * R_raw + c8 * G_raw + c9 * B_raw

where

c1, c2, c3
c4, c5, c6
c7, c8, c9

is color space transform matrix and depends on camera / filters and color space you are aiming to transform to. In general it is not easy thing to find this matrix, but you can do single star calibration, or you can solve for number of stars, or you can inspect QE curves and create transform, or you can take tool to do it for you - PI has star color calibration tool - use that compose RGB image out of mono images, do color calibration, split result again into mono images.

 

[celestron8g8]

That is a fantastic capture and edit ! Love that WF mosaic !

[MartinFransson]

12 minutes ago, celestron8g8 said:

That is a fantastic capture and edit ! Love that WF mosaic !

Thank you! I´m very pleased with it myself... might even print it  

[celestron8g8]

It’s definitely what i call a wall hanger ! What would be max print out size without degrading the pixels be ? Just curious . 

[MartinFransson]

36 minutes ago, celestron8g8 said:

It’s definitely what i call a wall hanger ! What would be max print out size without degrading the pixels be ? Just curious . 

The size is 6240*3333 pixels so that would be about 21*11 inches at 300 dpi, but for larger prints I find 200 or even 150 dpi is enough so that would be up to 42*22 inches. The thing is - the larger the print, the further away you watch it from  I have printed a 5 megapixel image at 3*2 meters and it looked OK since you had to stand far away to see it properly.

[Mr Spock]

That's a startling image. Seriously good.

[MartinFransson]

18 hours ago, Mr Spock said:

That's a startling image. Seriously good.

Thank you!