This article explains a practical approach to one of the difficulties of narrowband images: the colors of the stars. The usual combinations (SHO, HOO) produce uncalibrated colors for the stars. This issue is usually solved by shooting also in RGB to replace the narrowband, but this leads to 2 additional issues:
- The investment in additional integration time that could be used to increase the narrowband integration time.
- The stars tend to have higher FWHM, as a result of imaging them with a broader filter.
To solve this, I worked in a workflow that achieves «correct» star colors from the already gathered narrowband data. This is the result:
The result is not perfect, but quite close, and it is really noticeable the improvement on the FWHM. It is important to note also that this is an initial approach, so there’s room for improvement with additional processing.
This workflow is entirely done in PixInsight, but it can be traslated to any other image processing software. The image used in this article comes from one of my narrowband projects, the Wolf-Rayet star WR134 in Cygnus. You can see the final result at native resolution in this link.
RGB PCC: the goal
The following image is the result of the RGB data after PCC calibration. This is the goal for our workflow.
The graph provided by PCC shows how well is the resulting color balance.
It is not the best graph that I have seen, probably because I imaging using a stack of filters (WO VR1 + IDAS LPS-P2) that cut some of the incoming wavelenghts, specially in blue. In any case the colors look quite good for my taste, so I considered as a good reference.
Even when the possibilities of narrowband combinations are infinite, the most common way to combine narroband images are SHO and HOO.
Seems logical to develop the workflow around the HOO combination, as it is the simplest narrowband combination possible (only needs 2 filters), and the correspondance with the wavelenghts of the filters is quite logical.
The following image shows a HOO combination where Ha is mapped on red, and OIII is mapped to green and blue.
In general seems not far from a well calibrated stars, but unfortunatelly a closer look reveals some weird colors on certain stars:
Blue stars are too green, while orange ones have become completely red (they are lack of green). That means that the correct green channel weight is different for every of the individual stars, so no global weighting on the channels will fix the color balance.
Green WEIGHTING SURVEY
In a HOO combination, green channel is produced purely by the O filter, but as stated, there are infinite ways to combine narrowband data, so we can create different green channels mixing H and O. The question is, which percentage of O and H will produce the closest result to a RGB combination? and, would it be close enough to be correctly calibrated by the Photometric Color Calibration tool of Pixinsight?
The following serie is the result of different weights of Ha and OIII applied on the green channel, fixing Ha for red channel and OIII for blue channel. HHO and HOO combinations didn’t worked with PCC, so I chose a combination close to them.
As expected, the graphs of the narrowband combinations doesn’t match to the RGB one, but they are not as bad as I expected.
Anyway, the goal is to produce nice color stars, so a closer look to some representative stars (in terms of color range) will be enough to decide which combination works best.
It is possible to increase the saturation, to check the hue of the different stars. And will give us the idea of what’s going to happen when we push the data to the limits.
Blue stars have a green cast, no matter the combination, so this will be adressed in a further stage of the processing. For the red stars, something as O80H20 or O90H10 would give a nice result.
I checked another areas of the image (and processed all this combinations to the end) and chose this combination:
R = Ha
G = 80%OIII + 20%Ha
B = OIII
It is important to note that this combination is the one I chose as the best for my imaging system son may be not valid for other optical combinations. In that case I would suggest to perform the survey proposed above.
I use a quite simple workflow that uses three images as a reference:
- Nebula (NB_sn): A starless image of the nebula which we should process to taste.
- Stars Colors (RGB_stc): The resulting image of the explained in this article.
- Star Mask (RGB_stm): A starmask extracted from the Stars Colors image.
With this 3 images we will do the following:
- Mask NB_sn with RGB_stm
- Replace NB_sn (Masked) with RGB_stc
As NB_sn is masked, RGB_stc will only replace the pixels allowed by the mask RGB_stm. In other words, we mask and protect all the nebula and replace only the stars.
PREPARATION OF THE IMAGES
Let see how to prepare the images:
1. Process the narrowband channels until non linear. (Transferring a standard STF)
2. Starnet on every channel.
3. Combine narrowband channels to taste (Not necessarily the same as the Stars Colors image).
4. Process to taste.
5. Rename to NB_sn (NarrowBand_starnet)
This will be our nebula image.
This image is the reason of the workflow. It is the image that will add the correct color on the final image. The summary of the process is the following:
1. Combine the H and O channels as explained in this article. This combination is made with PixelMath.
2. Run Photometric Color Calibration on the narrowband image.
Note that PPC is operated in Broadband mode. It is suggested to use Background Neutralization using a preview placed on a background area of the image.
3. Convert the image to Non Linear transferring an standard STF.
This is the result:
4. Now use Curves (Color Saturation would work also) to increase saturation (quite hard).
4. As stated, the result is not perfect. This can be noticed mainly on the green cast remaining on the blue stars. To fix this we can apply SCNR to green. The standard parameters worked well on this image.
4. Now the image is ready. Rename to RGB_stc (RGB_STarColor)
This will be our stars color image
1. Extract Luminance from RGB_stc
2. Apply Starnet (starmask mode) to the luminance, to get a only-stars image.
3. High clip using the «darkest» star core as a reference and black clip to remove background.
4. Blur to taste to recover a gaussian shape on the stars (clip can make them too flat). I used ATWT removing the first layer.
6. Slightly dilate the stars. I found that this allows to include more color information from RGB_stc.
5. Rename as RGB_stm (RGB_STarMask).
This image will give the stars shape.
Note: Other tweaks can be desirable depending on the following result…
As a result we have 3 images named NB_sn, RGB_stc and RGB_stm. Now we can combine them.
1. Mask NB_sn with RGB_stm
2. Open PixelMath:
2.1 Check to use a single RGB/K expression
2.2 Type RGB_stc on the RGB/K channel
2.3 Uncheck Rescale
2.4 Check Replace Target
2.5 Apply to NB_sn
As a result, pixels allowed by the mask RGB_st will be replace by the RGB_stm on the NB_sn image.
As stated, additional processes can be applied to improve the result. I wanted to increase the color contribution of the stars, so I did the following:
1. Applied a very light erosion to the stars (using the RGB_st mask) using Morphological Transformation
2. Applied a second color saturation, keeping the mask active.
This is the result before and after this two additional steps:
And finally, the comparison between the RGB stars combination, and the Narrowband stars combination explained in this article.