Starless technique for Hubble Palette narrowband images
I have been refining a technique in PixInsight that mimics the tone mapping technique used in Photoshop, the basic stages of which are:
1. Removal of stars which is carried out after a non-linear stretch.
2. Combining the hydrogen alpha data as a luminance in the CieLab space.
3. Removal any gross artefacts arising from stages (1) and (2).
I am still developing these techniques and encourage you to experiment to see what works best for your data, workflow and personal tastes.
Linear Hubble Palette image
In the image above I have produced the initial SHO linear image by using PixelMath to combine the three narrowband stacked master files. In this instance IC 1871 "The Whirling Dervish Nebula" part of the Soul Nebula IC 1848.
1. I have unticked the option "Use a single RGB/K expression" as I want to assign the stacked narrowband master files to specific colour channels, ie SII to Red; Ha to Green; and OIII to Blue.
2. I have entered an identifier for the resultant image "Int_SHO".
3. I have changed the "Color Space" option to 'RGB Color".
Clicking the square button then runs the math expression and the three master narrowband stacks are combined.
The STF process has the channels set as "unlinked".
STF stretch of the linear SHO combined image
The next step is to stretch the image. Depending on the target's characteristics I will use either an iterative approach with the Histogram Transformation process or the transfer of the STF paramters to the Histogram Transformation tool which are then applied in one instance. I have found that if there is a strong signal in the image and abundant nebula, the STF transfer approach can be of benefit to optimse the detail in the final image.
I chose this technique in this image of IC 1871 as it has tremendous depth and signal strength in the nebulosity. You do need to exercise caution however as more subtle targets can easily be over stretched. I suggest you try to stretch your SHO image using both techniques and decide which you prefer or combine the two to produce a third blended image using PixelMath.
Removal of magenta halo around stars
It is important to rectify any anomolies with the stars at this stage as we do not want any artificial colours lingering in the image when we carry out the star removal process.
There is a straightforward three stage process to removing the magenta halo from the stars that arises from the channel combination.
1. Invert the image from the menu "Image>Invert". This then causes the magenta to display as green.
2. Apply SCNR set to Green using a figure in the "Amount" slider that removes the degree of green that you want. In my example I used 0.80, ie 80%.
3. Invert the inverted image to display the final result.
Inverted image after SCNR application
Resultant image after magenta removal
The image above is already beginning to look like a presentable image - which indeed it is. You could opt at this stage to forego the starless technique and proceed to further post-linear development and refinement of the image.
I have found that the additional steps taken to remove the stars and add a deconvolved luminance from the hydrogen alpha data does add much more depth and tighter stars.
The next steps involve the preparatory work necessary for the star removal with the creation of a star mask.
Star mask creation using the DSO Mask script
Open the DSO Mask script, "Script>DSOMask". Make sure that the correct image window is selected. If you find that the DSOMask script is not present in your script menu choice, you can add new scripts as they are developed by downloading them from the PixInsight repository.
Click "Create Mask" using the default settings.
DSOMask output image
The output image is an inverted image, ie the stars are black, and we need these to be white so the underlying stars are unprotected. Simply invert the image as before "Image>Invert".
The inverted star mask ready for refinement
The star mask needs to be stretched further to make the background darker for greater protection and the stars whiter to reduce protection. You achieve this simply by using the Histogram Transformation process to stretch the star mask. I select the "Auto clip shadows: %1.000" icon on the tool and apply this to the mask.
The finished star mask
We now apply the mask to the SHO image to protect the background and reveal the stars. Then using the Morphological Transformation process we iteratively reduce and virtually eliminate the stars from the SHO/RGB image.
Star reduction and removal 1
I have left the "reveal" mask option active so that you can see the red protected areas (black in the star mask) and white areas which overlie the stars in the SHO/RGB image. It is important to check the coverage of the star mask so that each star and its small halo is revealed by the mask. If the white stars in the mask are not large enough the reduction and removal of the stars will leave behind excessive artefacts; if too large, it is possible that underlying SHO/RGB data will be effected.
We apply the Morphological Transformation process in three separate steps:
1. "Amount" set to "1.00" (ie 100%) and "Iterations" to "5" as a first application.
2. The second application "Amount" set to "0.5" with "Iterations" left at "5".
3. The third application "Amount" set to "0.25" with "Iterations" left at "5".
Star reduction and removal 2
Star reduction and removal 3
Star reduction and removal 4
We are now approaching the final steps of the star removal and reduction and are now applying a further removal step that will achieve a minimal presence of stars. They will not be removed entirely (as a programme like Straton will do) but remarkably they are removed sufficiently that when we add the Ha stacked master there will be no discernible remnants.
In this step we carry out removal of layers leaving only the large scale layer within the space the star once occupied. We do this by using the MultiScaleMedianTransform process, increasing "Layers" to "6" and deselecting all of the layers except the "R" residual layer. Apply twice.
Star reduction and removal 5
After the last application of MMT, we now want to blur the residual star structure and we use the Convolution process to achieve this blurring.
You will need to increase the degree of convolution by adjusting the "StdDev" number - increase this to 30 and apply it to the star mask protected SHO/RGB image.
Star reduction and removal 6
Execute a further convolution step after incaresing the "StdDev" to "90".
Curves boost to starless image
At this point, I will often choose to boost the colour by subtle use of Curves: in this instance with both "Saturation" and the "C" curve.
Luminance extraction for use as a mask in later processing
Using either the ChannelExtraction icon in the menu bar or the process, extract the luminance from the starless SHO/RGB image. This image then needs to be strectched further and 'black-clipped' to form a very useful mask in later processing, eg as a mask for Local Contrast Enhancement.
Black-clipping the starless luminance to form a mask
Next we need to turn to the Ha master stack and prepare it for stretching and then combining as a luminance for the starless SHO/RGB image.
The Ha master image deconvolved and stretched
I have not detailed in this tutorial the deconvolution process (see my tutorial here). I stretched the linear Ha image using iterative steps with the Histogram Transformation process.
Before I combined the Ha with the starless SHO/RGB I carried out a small degree of star reduction.
Star reduction of the Ha
You can see that I applied a 20% reduction with two iterations.
Optionally at this stage you could carry out some light noise reduction or enhancement of the Ha image, eg the following processes HDRMultiscaleTransform or LocalHistogramTransformation. I choose to carry out these steps at a later stage to a separate cloned image and then blend the "enhanced" with the "unenhanced" to suit my preferences. I find this gives a greater degree of control.
Ha combination with starless SHO/RGB image
The culmination of this preparatory work is the combination of the Ha as a luminance in the CieLab colour space using the ChannelCombination process selecting "CieLab" as the "Color Space".
I trust you can see the greater depth to this image: HaSHO than the earlier stage when we removed the magenta halos after the STF-stretch of the SHO/RGB image.
Removal of chrominance artefacts with TGVDenoise
If you zoom in on the HaSHO image you will see tiny coloured remnants of the stars and also some faint halos around larger stars. Using TGVDenoise set to "CieLab" mode with "Lightness" deactivated and 'Chrominance" activated will remove these artefacts and repair halos without smoothing the lightness we have added with the Ha.
I use two applications, the first with 350 iterations, the second with 150 iterations. Note that "Local Support" is ticked with "Strength", 'Edge Protection" and "Smoothness" left at default values.
The image is now ready for further processing - such as Curves, SCNR and Colour Masks to modify the palette, HDR MulitScaleTransformation, contrast enhancement and noise reduction.