2015 - Updated PixInsight Monochrome Narrowband Workflow

There have been a number of changes in PixInsight over the past few months and together with reading others workflows has led me to review my own standard workflow for
narrowband images; mono Ha in particular. I thought it might be useful to share it.

As ever with PixInsight some degree of experimentation is key so don't take my exact settings as gospel, they do need tweaking due to conditions and the particular
characteristics of your setup. Mono image presentation is also open to a degree of personal style, I like a bit more contrast in my images than many. Hopefully, though, this will
provide a few ideas to some people but equally any builds on this are very welcome.

This is not a full PixInsight Tutorial, it assumes reasonable knowledge of the Pixinsight Interface

Chris Bailey - Nov 2015

ps I still do my calibration long-hand rather than using the batch pre-processing script.

1: Blink

Having backed up the base frames the first thing I do is Blink through the image set [Image Inspection>Blink] and MOVE any that don't come up to scratch to a REJECTED
folder. It is quite heart-breaking to cull long narrowband subs but it is important to be quite ruthless. If you have less than say 12 images, it is important to remove any with
satellite/plane trails as the tight rejection criteria needed to get rid of them from the stack can be detrimental to the final result. I remove any at this stage that have a visually
lower signal than the rest. Several images in this set were affected by passing bands of high cloud i.e. a pretty normal night!.
2. Image Calibration

It is important to get the tick boxes right on this one.
BIAS frame calibration should be left unticked. If you have not previously BIAS subtracted your DARK frames then calibration should be ticked and if your DARK frames don't
match the LIGHT frames exactly (duration and temperature) then tick the optimisation tickbox. I always BIAS subtract my FLAT frames prior to Flats integration so calibration of
the FLATS should be unticked as they have already been done.
NB with shorter subs on the SX814 Sony sensor I don't bother with DARK's but with 20 minute frames I do see some small benefit in using them though they are a pain to
collect in sufficient numbers
3a. Sub Frame Selector


I have only recently started using this. It is a bit time consuming but can spot 'dodgy' frames that are not necessarily visual apparent. It is very customisable!
The script measures all the input frames and gives a wealth of information (too much?) about them that may not be obvious visually. For me there are two important
parameters, FWHM and Eccentricity.
3b. Sub Frame Selector


The script enables the setting of approval parameters to reject all images that don't meet the criteria set.
Setting the rejection criteria can be quite sensitive but images can be manually ticked to approve them, if say a rejection criteria of <= 3.0 FWHM is set and one image has a
FWHM of 3.05 then I would manually approve that for inclusion.
Many people use FWHMSigma and EccentricitySigma in their expression settings but I find I have a better feel for straight FWHM and Eccentricity.
3c. Sub Frame Selector


Plots help visualise how the images vary within the criteria bounds set.
3d. Sub Frame Selector


Finally the script can be setup to mark images as approved or rejected and even move rejected images to a REJECTED folder.
4. Star Alignment


I have reverted to using a BiCubic Spline interpolation algorithm as I find Lanczos interpolation to create undershoot ringing on images taken with my SX814. With dithered sub
frames this is less of an issue but can still give some dark halos around brighter stars. Setting this to Auto will work for most people but some experimentation can be
worthwhile.
5. ImageIntegration


Having been pretty ruthless with image rejection, only 11 remain to be stacked. This is borderline for using a sigma clipping rejection but it does give better results with >10
images though a few more is ideal. I have set a permissive value for Sigma High as I know from blinking the images that there are no trails in any of the sub frames.
6. Crop


With all images it is worth cropping off a few pixels from the image edges. With dithered images this can sometimes need to be more than shown here.
7. Parameters for MURE De-Noise


The MURE De Noise Script needs values for the Sensor Gain and an estimate for Gaussian Noise. The former can be obtained from manufacturer data, the latter by putting a
couple of dark or bias frames into the [Script>Image Analysis>DarkBiasNoise Estimator] script. The author of both scripts is intending to change the nomenclature of both noise
settings to match up.
8. MURE De-Noise


This is currently not a part of the standard PI distribution so need to be downloaded from the link here http://pixinsight.com/forum/index.php?topic=9206.0 and then copied into
the PixInsight Scripts folder and added via [Script>FeatureScripts>Add]
MUREDeNoise seems to work very well on mono images with a reasonable signal to work with. It does not lead to the "painted" look of many other denoising algorithms and
does not lead to background clumpiness. Equally it is very good at preserving fine detail BUT does not remove every grain of noise.
It is an iterative process using the CCD and Noise parameters as a seed. 8 Spin Cycles works well but I see an improvement in upping this to 15. It is quite processor intensive
and on this 3300x2700 image, denoising takes about 5 minutes with 15 spin cycles.
9. Dynamic Background Extraction (DBE)


Placing samples manually is quite critical on many of these extended Ha objects. Only a few samples can still remove large scale gradients.
10. Masked Stretch


This had fallen out of favour with me until I tried reducing the target background and increasing the number of iterations significantly. I also reduce the clipping fraction by
several orders of magnitude.
Masked Stretch produces tighter stars than doing a normal histogram stretch. It is important to provide Masked Stretch with a background sample via a preview and setting the
Background Upper Limit value to the average background level sampled from within the preview box.
11. HDR Multiscale Transform


I like some contrast in my mono Ha images and it is very easy to overdo it with this tool. 7 or 8 layers with lightness mask selected produces a subtle increase in highlight
contrast, lifting detail.
12. ACD Noise Reduction


MUREDeNoise does a fine job of noise suppression but can leave some visible noise in the shadows. A gentle application of ACDNR with a very strong high signal area
protection mask can provide some further suppression in the darker areas of the image. Watch for ‘clumping’ in the lower signal areas and be prepared to back off the Amount
setting until this is minimised even if that is at the expense of additional noise in the shadows.
13. Local Histogram Equalisation


Another tool that can be a bit of a contrast sledgehammer but turned right down can lift overall contrast in a much gentler way than resorting to an S-Curve. When the preview
looks about right, I tend to turn the Amount setting down by a further 0.1! Use with an applied Luminance Mask for best results.
14. Histogram Transformation


I used to reach for the histogram tool early in the workflow but now prefer to make a single final adjustment near the end and aim to just clip the shadows by a little whilst
stretching the image slightly (to taste). The benefit of doing the second stretch at this stage is that the various contrast enhancement and second stage noise reductions will
have had an impact on the histogram and as the stretch can be data destructive it seems best to do it late on and only once.
15. Star Mask


It is often useful to do a round of star reduction to reduce the visual impact of the stars on the object being imaged, particularly in sky areas with dense star fields. Key to star
reduction is a good star mask. The settings for this tool take a degree of trial and error but the end goal is to have a mask that when applied and inverted, covers all but the
smallest stars. I find it useful in star reduction to reduce the smoothness value from the default and increase the scale and noise threshold to grab the larger stars and avoid
background noise respectively.

Don't forget to invert the mask back before applying star reduction.
16. Morphological Transformation (Star Reduction)


This is a tool I have spent several hours playing with. I used to use Morphological Erosion for star reduction but found it to produce stars with am un-natural hard edge profile.
Morphological Selection does a better job but has two sliders to adjust. I also find it works best with a multi-way circular structuring element. The default is one way but by
clicking on the +[Add Way] it is possible to draw in the box to produce a much more believable stellar profile.

The end result is a very subtle star reduction.

[Script>Utilities>DarkStructureEnhance]. This one is very optional and I don't use it on all images but, used sparingly, it can add further to an images shadow contrast and with
an object like this it can be very useful in bringing forwards fine shadow detail.

Multiple iterations with a small Amount Setting tends to give better results than a single large dose.
19. Unsharp Mask


Again, very easy to overdo it. Apply with a L mask and turn down the settings. Undo>Redo to make sure it is not overcooked and be prepared to turn down further. The impact
of sharpening should only really be discernible at 100% zoom levels.
20. Halo Reduction


On bright stars I get a reflection back of the Tak rear element. Its nigh on impossible to get rid of them completely but this does reduce their impact a little.
Use PixelMath to create an ellipse centred on the halo and with the dimensions set to cover it completely and create a new image
Convolve the result significantly and apply as a mask
Using curves, reduce the brightness of the halo. You cant overdo it or it becomes obvious.
21. Integer Resample


For posting to UKAI it is necessary to downsample an image. Doing it by an integer amount tends to give better results than downsampling to a fixed scale
22. Completion