BlurXterminator

[PRejto]

Hello,


I've been experimenting with BlurX and have been pretty impressed with the results.



Today on an image I noticed that the smallest stars were being reduced to a pixel size where the stars have a square appearance even with quite modest settings. It isn't noticeable at 1:1 viewing but becomes quite noticeable at higher magnifications. Larger stars look great and detail pops in the image. I imagine I could try to mast the smallest stars but I've yet to run into anyone using a mask with BlurX to do this.


I should add, this is from my TEC180 with a .9 FRC and a KAF-8300 CCD camera giving 1 Arc-sec resolution. I wouldn't say the data is under sampled.




Would you ignore this issue? Thanks,


Peter

[Startrek]

Peter,
Great to see folk experimenting with new processing concepts ( however BlurX is not my cup of tea )
Not sure if image sampling affects BlurX in relation to stars , however I use Startools Spatially Variant PSF Deconvolution to reverse atmospheric blurring in my images and sampling plays a significant factor when it comes to keeping all stars but in particular those smaller stars “roundish” rather than “blocky” when zooming in or pixel peeping.

My sampling depending on which scope I use is between 0.62 and 0.88 arc sec per pixel ( oversampled) which allows me to software Bin my data to a specific image size which best suits SV PSF Deconvolution.
Binning in Startools is scalable doing I can choose what level as a percentage works best. It’s not a set bin like 2x2 or 3x3 or 4x4 etc.

https://www.startools.org/modules/sv-decon

https://www.startools.org/modules/bin

If SV Decon is applied with the most suitable image size ( Binned as required) then generally I can zoom in to 300% or more and small stars still retain circular / centroid shape. If I Bin to aggressively to trade some resolution for noise reduction and the image size is too small then SV Decon will result in the smaller stars being a bit blocky looking.

Here’s an image from last year M16 which shows before and after SV Decon is applied for atmospheric de blurring
Unfortunately IIS reduces image to 200KB size so resolution is greatly reduced but you can compare before and after images

Cheers
Martin

[PRejto]

Thanks for your detailed reply, Martin. I don't know StarTools but perhaps I will investigate!


According to Russ Croman he considers my setup to be borderline under sampled in 2 arc-sec seeing and the blocky looking small stars to be expected. He suggested enlarging the halo. This helps a bit but seems to defeat the purpose on larger stars where the tool seems to operate perfectly.



I may try making a mask only for the very tiny stars and see what happens.
Just how I make that mask may be the challenge as I'm pretty new to Pixinsight.


Peter

[Nikolas]

That's assuming one uses startools Martin.
I am no fan of pixinsight however because of Blurxterminator I bit the bullet.

Blurx will always work at its best if you have decent data. I experimented with some older data and if there wasnt enough acquisition time it wasnt the best but data with decent signal actually was incredible with blurx then noisex the images are pristine.

I experimented with startools and just didnt like it but each to their own, every program has its good and bad.
Cheers

[Nikolas]

Peter if you get blocky stars you are undersampled, a workaround is to drizzle x2 your original stacking then you use resample tool in pixinsight, this eliminates the blockiness but does add time to your processing.

[PRejto]

Nik,


Is it worth it? I'm here only speaking of the very tiny tiny stars. If you look at the first image I posted starting this thread does that image look under-sampled? I see the blocky core but a rather round halo.


I did make a mask that will protect those very small stars, but using it and taking full advantage of the tool gets complicated, meaning star removal in order to process the background, etc.


P

[alpal]

I might get it if it ever works directly with Photoshop as a plugin.

I couldn't bothered learning PixInsight even though I should.
Imagine having to buy PixInsight and tear your hair out for months
trying to use it just to be able to use BlurXTerminator?

cheers
Allan

[Emuhead]

I've found using AI version 1 works well so would recommend trying that. Huge fan of BlurXterminator myself. There's a YouTube video of Russell Crowman talking with Adam Block on every detail, worth a watch.

[Shiraz]

Hi Peter.

If its good enough for Hubble, maybe Ok for you?

attached is part of a (blown up) full res Hubble image showing star shapes. Just like yours...look good to me.

As Nik already noted, after deconvolution, you have symptoms of slight undersampling, but I would just rejoice in the tiny stars and leave it at that.

Cheers Ray

[Craig_]

I run StarXterminator on the stacked, linear data then BlurXterminator on that file. Thus, it does not impact the stars, and you can then add the stars back in easily later.

[Craig_]

You don't really need to know how to use PI to use BlurX.


Bring your linear data in from another program and run BlurX, then save it as linear data and send back to your program of choice. Running the BlurX script itself is very easy and does not require a working knowledge of PixInsight.

[strongmanmike]

Thats what I was assuming, people are claiming "oh its not that easy, I do lots of clever things blah blah blah" :rolleyes: :lol: I kinda recon its more about forking out the dosh $ for PI aaand BEx plugin and bingo, all our images look awesome aaand the same..a playing field leveller if I have ever seen one....as Ive said, like steroids :lol:

Astroimaging trials and tribulations...:lol:

Mike

[Geoff45]

Horses for courses. I found that learning PixIsight was one of the most fun things I ever did in image processing.

[alpal]

Sorry Guys - I wouldn't know,
I have only heard stories of PixInsight having a steep learning curve
requiring months of work to use properly.
There are 10 pages on it here:
https://www.cloudynights.com/topic/616865-post-your-pixinsight-processing-flow/


cheers
Allan

[Nikolas]

Blurx Starx and Noisex has simplified things in PI along with Bob's and Paulyman's scripts

[Craig_]

You can get the free 30 day trial for both PI and BlurX.


It's literally this easy (if you have a stacked file from other software):


1) Open linear stack
2) Process (top menu) > All processes > StarXterminator
3) Press Square (Apply) on the StarX process window
4) Wait
5) Once done, Process > All processes > BlurXterminator
6) Adjust settings to taste, some are irrelevant having removed the stars, all I change now is the sharpen non-stellar field I run at around 0.60 or so, which works well on linear, starless data.
7) Press square (apply) on BlurX window
8) Wait
9) File > Save as
10) Open in whatever other tool you prefer to stretch and process, and add the stars back in from the original linear data stack


To be fair my steps above don't include the installation of either plugin but this is also quite easy and instructions are provided. BX is well worth learning.

[alpal]

What if you wanted Blurxterminator to fix up bloated stars?

[Craig_]

I haven't used it for that, but if you wanted to use it on the stars just skip my StarXterminator steps and play with the BX settings.


Depending on the nature of your star bloat you may have more luck with other methods though.


Create a starless layer, then put the original (with stars) layer on top of it (in Photoshop, or your program of choice), set the blend mode of this star layer to screen, then create a curves adjustment layer and pull down hard on the curve. Convert this to a clipping mask. Then create a brightness/contrast layer, reduce brightness significantly (how much will vary by image), and convert to clipping mask.



You should now have significantly smaller stars. Play with the intensity of the two clipping masks to taste. This technique also works well for blending in RGB stars to narrowband images.

[alpal]

There are many methods:
https://www.youtube.com/watch?v=qSnIvJru5Wo


Louie had a good one there.

[RobF]

Hi guys, I shared these pics on the local FB astro group. What started as an afternoon of determined deconvolution research ended up in amazement when gave the BXT trial a spin (in PixInsight).

Should also add the although not alway intuitive for some people, PI was a breath of fresh air for me over 10 years ago - and still gives a lot of satisfaction for hard won data.

This FOV is a tight crop. System used was FSQ106ED@F5,
QHY247C (1.52 arcsec/px) - Approx 100mins of 1, 3, 5min subs.

Definitely not oversampled - so definitely pleasantly surprised to see some benefit. $100US not cheap, but not hard to blow that much money on a USB hub in this hobby......

[Spookyer]

Hi Rob, very nice results there.

Brett

[ChrisV]

That's really nice Rob. I've often skipped deconvolution as I couldn't get much sharpening without weirdo stars. Tried piles of parameters and masks ... Grrrr. Resorted to a light sharpening instrad.

Might give the blurX trial a go. I could do with a bit of artificial intelligence

[RobF]

Thanks Brett and Chris. I should have also said I went a bit psycho - from memory there were a couple of runs of BXT, some curves and possibly some PI Dark Structure Enhance script.
Understand what you say about deconv Chris - I've agonised over masks etc and found it so hard to get anything decent - and even then its often so harsh has to be toned down.
I don't think its inventing data - seems plausible what it ended up with, but gather that's one of the concerns/arguments with AI driven tools.

[joshman]

BXT is a fantastic bit of software. And i've found some good settings that work for me:

Sharpen Stars (+0.5)
Adjust Star Halos (+0.2)

Sharpen non-stellar (0.6)


I find these to work pretty good if i'm blending RGB stars into a Narrowband image.

I've also foudn that you need to be a bit careful with BXT, as it can sometimes squash any subtle details in very dark areas.

[marc4darkskies]

I agree wholeheartedly, BlurX is a great tool. Thankfully, one doesn't have to delve too deeply into the the acronym maze that is Pixinsight to use it! I basically do a histogram stretch before applying BlurX and then export the result into PS where I can blend the result to taste.

As with all sharpening though, great care must be taken to ensure that the image doesn't look overcooked. With respect to your example image, and notwithstanding it's essentially a test, I'll be brutally honest and say it IS somewhat overdone. By that I mean it's obviously been sharpened - a lot. If this happens to me, I dial it back either by changing parameters or by blending it with the unsharpened layers below (my preferred method)

Anyway, a good, informative post Rob with an decent result.

Cheers,
Marcus

[RobF]

Thanks for that extra info Josh and Marcus. Totally agree this extreme crop and 2 test applications was overdone, but glad to hear many are finding these RC algorithms of use. Also found this write up interesting, including the frank break down of arguments against such tools:

https://cosgrovescosmos.com/tips-n-techniques/blurxtermintor-a-breakthrough-for-decon?fbclid=IwAR0ErIGc7SjCMYl0avcE a35H4IVCCV6ll7QysH0k258V_CDOJ4OcmYj ehNA

[RobF]

Disclaimer: Robo astrosystems gets no financial kickbacks from RC Astro! :)

V4 of BXT now allows it to be run on stars only and significantly improve coma, star shapes etc with variable correction across FOV. Luminance only mode also makes it easier to run on non-stellar nebs etc without affecting star colour.

In PixInsight, my workflow has changed substantially:

WBPP (has improved out of sight in recent years)
BXT for star correction only
ABE/DBE if required for gradients
SPCC for color calibration/correction
NoiseXTerminator (if you gone all in for RC tools)
BXT (deconv)
Then stretch, or do a star/starless workflow, etc etc

In cloudy times have had immense fun with modern PC firepower and just revisiting the astroimage archives re-working old images - particularly those where there was lots of data and could never get it to play nicely.

What a great time to be an astrophotographer.
Wait, when there's no clouds, what a great time to be an AP'er.... ;)

[ChrisV]

Thanks for the info. I've just started using it and have been doing bXt after DBE. Do you think it's important to do it before DBE?

[RobF]

Hi Chris

From playing on my data and trawling threads don't think anything is set in stone. Russ Croman seems to recommend BXT best on linear data (particularly if being used for deconvolution versus working on stars alone). There's a big argument on PI forums about if BXT for star shape correction before SPCC is worthwhile. Likewise some say ABE/DBE before SPCC, others prefer doing on stretched data.

The one universal seems to be more time spent in the field getting lots of well tracked and calibrated data makes everything later in processing easier.

Doesn't cost anything to save your masters and working images as you progress in a folder, and later try doing things differntly to see if any better or worse.

Sorry if that doesn't answer the question :)

Would be interested in any insights others have or if you've adjusted your (PI) workflow over last year or so? With apologies to anyone preferring Startools, Photoshop etc......