Producing 'RGB' stars from Ha, O3 and S2 data

When we are doing narrowband imaging we frequently don’t have time to do separate RGB imaging for the stars. This is due to a number of reasons:

The fact that clear nights are so infrequent.

At our imaging place there is only a relatively short window of opportunity to capture narrowband data on any target before it runs into obstructions.

As our imaging is exclusively for the testing of Nicola’s capture software AstroDMx Capture, it is in the best interests of our imaging to capture as much data as possible through each narrowband filter.

Thus the fact remains that we frequently have to use less than ideal stars in terms of their colours. Frequently we have used HOO data to obtain a stars image and then gently adjusted hue and saturation to produce subtle star colours that are usable in the final images.

There are, however  pixelmath based procedures that produces more realistic star colours.

Some pixelmath formulae are:

Method 1

R = Ha * 0.8 + SII * 0.2

G = OIII

B = OIII

Method 2

R = 0.4 * Ha + 0.6 * S2

G = 0.4 * O3 + 0.3 * Ha + 0.3 * S2

B = O3

A simpler blend using only Ha and O3 data:

Method 3

R = Ha

G = 0.2 * Ha + 0.8 * O3

B = O3

These methods produce different but acceptable star colours.

It seems logical to combine the three methods into an average pixelmath formula:

Blend of the three methods (Method 4):

R = ((Ha * 0.8 + S2 * 0.2) + (0.4 * Ha + 0.6 * S2) + (Ha)) /3

G = ((O3) + (0.4 * O3 + 0.3 * Ha + 0.3 * S2) + (0.2 * Ha + 0.8 * O3)) / 3

B = ((O3) + (O3) + (O3)) / 3

Which simplifies to:

R = ((Ha * 0.8 + S2 * 0.2) + (0.4 * Ha + 0.6 * S2) + (Ha)) /3

G = ((O3) + (0.4 * O3 + 0.3 * Ha + 0.3 * S2) + (0.2 * Ha + 0.8 * O3)) / 3

B = O3

The data used here are of the Tadpoles nebula IC410, LBN807. Captured by AstroDMx Capture with a QHY Minicam8 through a William Optics 81mm APO refractor with a 0.8 reducer/flattener. The Ha, O3 and S2 data were stacked and part processed in PixInsight. The stars were removed and kept as Ha, O3 and S2 star images. The starless images were processed as described in the previous article to produce the Gendler palette which for the tests here was channel shifted to Gendler-GRB. The stars were then processed in PixInsight with BlurXterminator set to correct only.

Click on an image to get a closer view

Starless image of the Tadpoles nebula in the Gendler palette RGB

Starless image of the Tadpoles nebula in the Gendler palette channel-shifted to GRB 

Gendler GRB with Method 1 stars

Gendler GRB with Method 2 stars

Gendler GRB with Method 3 stars

Gendler GRB with Method 4 stars (pixelmath average of Methods 1,2 and 3)

Each method produces stars that are subtly different, particularly the redness of the red stars. Methods 1 and 3 produce vivid red stars whereas in Method 2 the red stars are a more yellowy red. The average of methods 1,2 and 3 produces red stars with a more gentle red rather than a deep red.

These are all subjective views of the results which are only an approximation to true RGB stars. However, the methods produce more realistic stars than any narrowband palette does and the stars are processed separately from the nebula. Our personal preference is Method 4 but in the end it is a matter of personal taste. Moreover, the methods presented here are not exhaustive and it is entirely possible that even more realistic star colours could be produced by other means.

Manipulation of narrowband palettes

Although a lot of the work done here could possibly be done in a single application, I find the most comfortable workflow involves the use of a number of programs. Most of the work will be done here in GIMP 3 but pixelmath will be done in Siril. Stacking, de-noising star removal and de-linearising were done in PixInsight.

Creation of a palette from grayscale images captured through Ha, O3 and S2 narrowband filters.

The data used here are of the Tadpoles nebula IC410, LBN807. Captured by AstroDMx Capture with a QHY Minicam8 through a William Optics 81mm APO refractor with a 0.8 reducer/flattener. The Ha, O3 and S2 data were stacked and part processed in PixInsight.

The monochrome stacked images have their stars removed and are de-noised. Then the starless images are stretched with Curves and then Levels in GIMP 3 so that their histograms are similar and have the same means as closely as possible. This gives the images similar ranges of brightness from darkest to lightest,

H-alpha image

O3 image

S2 image

Producing an SHO image by mapping the S2, Ha and O3 images to R, G and B

Assigning monochrome images to RGB channels

Produces the resulting SHO image of the Tadpoles nebula

The stars can be returned by pasting the stars image onto the starless images and select Screen mode. With the pasted layer in focus, Curves can be used to adjust their prominence to the desired degree before flattening the image.

The other seven narrowband palettes: OHS, OSH, HOS, HSO, SOH, HOO and SOO can be made by the same method in GIMP 3.

There is another group of narrowband palettes that are produced by pixelmath and that was done here in Siril. These palettes are the Gendler palette, the ForaaX palette, the Natural Palette and pixelmath variations on these.

The Gendler palette constructed by pixelmath in Siril

The Gendler palette of the Tadpoles nebula RGB rendering

The stars can be added back as before

Because each channel is a complex pixelmath construct, another process called channel shifting is used to contruct  a further five palette renderings from a palette such as the Gendler palette. 

The Gendler palette image is decomposed into its RGB channels as layers

Decomposing to layers

Decomposed image with the three monochrome layers shown at the right hand side of the screen

The RGB channels are re-mapped in a different order to BRG  (channel-shifting)

The RGB image has been channel-shifted to BRG a totally different rendering

By this procedure, it is possible to produce all five of the channel-shifted renderings of the Gendler palette in addition to the original RGB rendering. The stars are Screened into the image as before, and their prominence adjusted with Curves before flattening the image.

Gendler palette BRG rendering

Gendler palette GBR rendering

Gendler palette BGR rendering

Gendler palette RBG rendering

Gendler palette GRB rendering

It is, of course, possible to similarly construct channel-shifted variations of any pixelmath generated palette.

It is no surprise that the six basic narrowband palettes: HOS, OHS, HSO, SHO, SOH and OSH are simply channel-shifted versions of each other, although they are more conveniently constructed by channel combination from the original monochrome Ha, S2 and O3 images.

It is worth exploring the channel-shifted variations of any pixelmath generated palette as they each reveal the structure of the nebulosity in different ways and some may be more aesthetically pleasing than others.

Using the starrem star-removal command-line software on Seestar stacked images

Command-line software comprises programs that are run from a terminal window rather than from a GUI (graphical user interface). This is reminiscent of using DOS or CPM or mainframe operating systems before the days of Windows-like GUI operating systems. In fact, Linux and even Windows users still use the command-line interface to achieve very powerful results. 

Location of the star-removal software:

https://github.com/code2k13/starreduction

For Linux a tar.gz file is downloaded:

starrem2k13_ubuntu_20.04.tar.gz

This file contains all of the components including the Linux binary

For Windows a .zip file is downloaded

starrem2k13_win.zip

This file contains a .exe file which installs the Windows binary and the other components.

The Linux and Windows command-line programs are used in almost identical ways. In each case a directory (folder) exists containing all of the required files.

In both operating systems the starry image is placed in the same folder. If it is given a short filename it will be simpler to use. Moreover, one should never use spaces in filenames in Linux. It is better practice to use an underscore rather than a space in a filename in any OS.

In Linux the executable file is called starrem2k13

In Windows it is called removestars.exe

In both Linux and Windows a terminal is launched inside the same folder as the executable and the starry image file (which in the case here we have called EastVeil.jpg). 

The following line is typed or pasted into the terminal of Linux or Windows.

The two lines are similar but not identical. The difference arises because the start of the line begins with two characters which are required by a POSIX environment (Linux) or a Windows environment. The other difference is in the name of the executable: starrem2k13 in Linux and removestars.exe in Windows.

Command-line Linux

./starrem2k13 EastVeil.jpg starless.jpg

Command-line Windows

.\removestars.exe EastVeil.jpg starless.jpg

When enter is pressed after the command-line has been entered, the stars will be removed from the starry image and a starless image (in this case starless.jpg) will appear in the same folder. This is somewhat reminiscent of the use of the command-line version of Starnet++ but has less stringent requirements of the file having the stars removed.

Windows

Screenshot of the star removal software running at the command-line in Windows

Linux

Screenshot while the star removal software is running

Screenshot after the star removal software has completed the star removal

Notice that the starless image has now appeared in the folder

It should be noted that there seems to have been no work done on this open source project since 2023. For macOS there only seems to be a Unix executable for x86_64 but not for Apple silicon. Readers can try the macOS version if they have an appropriate machine but we shall look only at the Linux and Windows versions. Also the author of the software has apparently produced some sort of Python front end, but we shall only be considering the command-line implementations.

Initial testing indicates that the software works with 8 bit and 16 bit image files but not with 32 bit images.

There is an online demonstration version of the software that runs entirely in your browser and allows uploading a square image of up to 1024 x 1024 and downloading the starless image at a lower resolution. Uploading an oblong image will allow star removal but the starless image is distorted into a square image.

https://ashishware.com/static/star_removal/index.html

Screenshot of the browser in which the demonstration of the software removing stars from an uploaded image and producing a starless downloadable image.

This would not be useful for a workflow but is intended for a potential user to judge the efficacy of the software.

The AI model was trained on data created by a GAN (Generative Adversarial Network) which is a powerful deep learning model using two competing neural networks: a Generator and a Discriminator to create realistic, new data. This is a process called data Augmentation: Producing more training data for other AI models such as the star removal software.

We are introducing this star removal software here for a specific reason.

It is noticed that a large number of smart telescope users such as Seestar users are content to display and share the results of live stacking with little, if any processing. If attempts are made to enhance the images produced by the Seestar, brightening the nebula will result in a concomitant brightening of the stars which is an undesirable effect, bloating the stars and tending to saturate them, losing the star colours in the process.

On 27 August 2025 I published a blog article on 'Using AstroEdit on iOS to process Seestar images'. That software gives iOS users the ability to remove stars from their Seestar images, brighten up the nebulosity to an extent and then replace the stars in the image. The problem is that at the time of writing, AstroEdit is not available in a fully functional form for use on Android devices or Amazon Fire devices at all, both of which are also used for controlling Seestars and capturing deep sky images.

If data are downloaded from a Seestar onto a computer, all sorts of possibilities appear. It is possible to stack all of the individual sub-frames in software such as PixInsight or Siril and to completely process the images in those programs as well as others such a GraXpert, Gimp, and many others. However, we are considering here the case when a user is content to useLocation of the star-removal software: for example the high quality JPG that is produced of the stacked image by the Seestar but would like to be able to boost the nebulosity a little without burning out the stars.

In such as case, just Gimp and the starrem star removal software can achieve good results without the huge learning curves associated with more sophisticated software. Also these programs are open source and free to ‘own’ and use.

It is conventional wisdom that one does not use JPG images for processing because the image is saved in a lossy format in which some information is discarded. In high quality JPGs it is still possible to do a limited amount of processing but it is best to save the results in an uncompressed format such as a PNG or a TIF file. It is by taking advantage of this fact that Seestar users can improve their images that were saved as JPGs by the Seestar.

So for example, one could use the command-line star-removal software to accept a JPG and to output a PNG or TIF. For example:

In Linux

./starrem2k13 EastVeil.jpg starless.png

In Windows

.\removestars.exe EastVeil.jpg starless.png

Starry JPG on the left and starless PNG on the right

In Gimp is a simple matter from the starry and starless images to create a star image which can be added back to the enhanced nebulosity image with any required level of prominence.

The starless image is pasted as new layer onto the starry image with a Mode of Subtract or Difference (whichever gives the best result). The star image is flattened and pasted as a new layer back onto the starless image which may have been enhanced with Curves and/or Levels. The Mode would be Screen or Addition (whichever gives the best result). Then with focus on the pasted star layer, Curves are used to increase or decrease the prominence of the stars as required. The image is then flattened and exported as the final processed image.

Glass moulds distant light

Pixels map ancient echoes

Cosmos shows its face

The East Veil Nebula with an SV705C OSC

AstroDMx Capture was used with an SV705C OSC 12 bit, uncooled colour CMOS camera and a StellaMira 66mm APO doublet refractor with field flattener and a ZWO EAF. An Altair 2” magnetic filter holder version 2 containing an LPRO-MAX filter was placed in the optical train.

The equipment as monitored by a network camera

The equipment was mounted on a Celestron AVX GOTO mount. An SVBONY SV165 guide-scope fitted with a QHY-5II-M guide camera was mounted on the imaging scope. An INDI server was running on the imaging Linux computer indoors. The guide camera was connected by USB to another Linux computer indoors running PHD2 autoguiding software via the INDI server. The mount was controlled by AstroDMx Capture via the INDI server and the focuser with a hand controller.

Screenshot of PHD2 auto-guiding

AstroDMx Capture slewed the scope to the star Deneb and plate-solved to centre it. A Bahtinov mask was used to enable Deneb to be brought into sharp focus.

95 minutes worth of RAW 5 minutes exposures were capture along with dark-frames, Flat-frames, Drak-Flats and Bias frames.

The data were stacked and part processed in PixInsight and further processed in GraXpert, Siril, SetiAstroSuitePro and Gimp

Screenshot of AstroDMx Capture saving RAW data using live-stacking to improve the preview.

Negative preview

The East Veil nebula

A PixInsight Cosmic Photons script was used to construct a Herzprung Russel diagram of the stars in the field of view.

This test was done specifically to test the Sony IMX585 CMOS sensor with the StellaMira scope in preparation for possible future implementation of equipment using a variation of this sensor.

PhotoDemon; free, open source Photo editing software

Every so often I come across some software that, as far as I am concerned, has been hiding in plain sight. Such a program is PhotoDemon.

PhotoDemon is developed by Tanner Helland.

It was first released in 2012. It is open source and freeware, although donations can be made to the project. At the time of writing PhotoDemon is on version 2025.4.

PhotoDemon is not installed, it is an executable along with its required dependencies and requires no administrator privileges. Aside from a temporary folder – which you can specify in the Tools > Options menu – PhotoDemon leaves no trace on the hard drive. PhotoDemon can be run  from a USB stick or SD card.

It is lightweight, fast, completely portable and works with 8 bit, 16 bit and 32 bit images; unlike many photo editing programs which reduce higher bit depth images to 8 bits before they can be used.

Due to its portable nature, PhotoDemon is only available as a 32-bit application. This means it cannot load or save images larger than  about 2 GB in size.

Complex editing actions can be recorded as macros. A built-in batch processor lets you apply macros to entire folders of images. 32 bit 8bf plugins can be installed and used. The 32 bit 8bf, Hasta La Vista, Green, HLVG, from DeepSkyColours installs and works perfectly in PhotoDemon.

PhotoDemon is very intuitive and usable. Small touches like real-time effect previews, save/load presets on all tools, unlimited Undo/Redo, customizable hotkeys, mouse wheel and X-button support, and descriptive icons make it fast and easy to use. For those who know something about Photoshop, many of the keyboard shortcuts work the same way.

PhotoDemon has very extensive file format support, including Adobe Photoshop (PSD), Corel PaintShop Pro (PSP), GIMP (XCF), and major camera RAW formats, but at the time of writing, does not support the Fits file format.

Advanced multi-layer support, including editable text layers and non-destructive layer modifications

Colour-managed workflow includes support for embedded ICC profiles.

Its on-canvas tools include digital paintbrushes, clone and pattern brushes, advanced selection tools, interactive gradients, and more.

Adjustment tools include levels, curves, HDR, shadow/highlight recovery, white balance, and many more.

Filters and effects include perspective correction, edge enhancement, noise removal, content-aware fill called ‘Heal selected area’, and resize, unsharp masking, gradient and palette mapping, and many more. More than 200 tools are provided in the current build.

Screenshot of PhotoDemon using Curves on a Star layer that has been Screen blended with the starless layer.

Screenshot showing the adjusted stars layer

PhotoDemon is right up there with Photoshop, Gimp and Affinity Photo. It will be one of the tools in my workflow for processing astronomical images. I can recommend it to any astrophotographer who wishes to use a powerful yet intuitive program to contribute to the processing of her/his images.

I was unable to get it to work in Wine in Fedora or Ubuntu Linux. If it can be made to work in Wine a special Wine ‘bottle’ may have to be developed.

PhotoDemon can be downloaded from https://photodemon.org/

The Lobster claw nebula region

AstroDMx Capture was used with an SV605CC OSC colour CMOS camera and an Altair Starwave ASCENT 60ED doublet refractor with 0.8 reducer/flattener and a Pegasus Focuscube v2. An Altair 2” magnetic filter holder version 2 containing an Altair quadband filter was placed in the optical train.

The equipment was mounted on a Celestron AVX GOTO mount. An SVBONY SV165 guide-scope fitted with a QHY-5II-M guide camera was mounted on the imaging scope. An INDI server was running on the imaging Linux computer indoors. The guide camera was connected by USB to another Linux computer indoors running PHD2 autoguiding software via the INDI server. The mount and the focuser were controlled by AstroDMx Capture via the INDI server.

AstroDMx Capture slewed the scope to the star Altair and plate-solved to centre it. A Bahtinov mask was used to enable Altair to be brought into sharp focus.

Screenshot of PHD2 autoguiding

Then AstroDMx Capture plate-solved and sent the scope to the star HD 240247. This star was chosen so that the image would frame a region containing five objects: SH2-157, the Lobster claw nebula, NGC 7635, the Bubble nebula, NGC 7538 (emission nebula), NGC 7510 and M52 (open clusters).

AstroDMx Capture was used to capture 26 x 5-minute RAW exposures of the region.

Screenshot of AstroDMx Capture saving RAW FITS files. Live stacking was used to improve the preview of the captured image.

Negative preview

26 light frames, 5 dark frames, 50 bias frames, 50 flatfields and 50 dark flats. To capture the flats and other calibration frames, the scope was slewed to ther zenith and tracking stopped. The flats were captured using a device designed to avoid lateral light leakage and using a stack of 3mm thick acrylic white A5 sheets to evenly diffuse the light from a light panel and attenuate it sufficiently to get Flats of 1.5s duration.

The Flat field equipment

The data were calibrated, debayered, stacked and part processed in PixInsight and completed in GraXpert, Seti Astro Suite Pro Cosmic Clarity and Gimp. A Pixinsight Cosmic Photons script was used to Generate a Hertzprung Russel diagram of the stars in the image.

Hertzprung Russel diagram of the stars in the stacked image

The Lobster claw nebula region. All images with RGB stars.

RGB channels linked

RGB blend of linked and unlinked channels

Annotated image

HOO rendering

Hubble palette rendering

Classical SHO rendering

Meanwhile Nicola's mammoth refactoring of AstroDMx Capture is progressing nicely, although with the decision to make a number of functional and efficiency changes in addition to making the software Wayland compliant with Qt6, the process has taken longer than was initially anticipated. The ground is being prepared for the introduction of new frameworks such as Alpaca as well as new functionality.