The Smokin' Hot Galaxy - M82 returns - this time properly smokin'

[ollypenrice]

1 hour ago, wimvb said:

Olly, the days of the Spanish Inquisition ended when the CloneStamp tool was implemented in PixInsight. 😉

Really? I think they should burn themselves at the stake in that case. They'll be introducing Layers, next, so you can see what you are doing while you do it.

[The Lazy Astronomer]

43 minutes ago, ollypenrice said:

Really? I think they should burn themselves at the stake in that case. They'll be introducing Layers, next, so you can see what you are doing while you do it.

Where's the fun in that - wouldn't want to make it too easy now, would we?! 😁

[powerlord]

1 hour ago, ollypenrice said:

Really? I think they should burn themselves at the stake in that case. They'll be introducing Layers, next, so you can see what you are doing while you do it.

that's crazy talk.

[ollypenrice]

20 minutes ago, powerlord said:

that's crazy talk.

The only talk I've ever known! 

Olly

[wimvb]

2 hours ago, ollypenrice said:

Really? I think they should burn themselves at the stake in that case. They'll be introducing Layers, next, so you can see what you are doing while you do it.

Sorry. Can't resist. Should I grab my coat?

https://youtu.be/aJQmVZSAqlc

 

[ollypenrice]

6 minutes ago, wimvb said:

Sorry. Can't resist. Should I grab my coat?

https://youtu.be/aJQmVZSAqlc

 

You might consider it!!!

[wimvb]

When I look at the deconvolved image presented in this thread, and its comparison to Hubble data, I don't see invented detail. The structures in the deconvolved image are similar to those in the Hubble image. But I think that the deconvolutuon has been pushed too far, introducing artefacts that would most likely also have been introduced by classic deconvolution. As Russel Croman states it so well on his web site, any deconvolution involves guessing.

When you see piles of sand on a beach, you can only guess what the sand castle that stood there a few hours earlier may have looked like. And with some detective work, one might even attempt a recreation of it. But such a recreation can only work up to a certain point. The equivalent in astro images should never be pushed beyond that point, or artefacts will be introduced. As a matter of fact, in the original PixInsight deconvolution process, much work went into adjusting parameters so that artefacts were not introduced, while at the same time as much detail as possible was restored. What I see in this excellent image of the Cigar galaxy, when enlarged, is just some of the artefacts that can be the result of any deconvolution. Some tutorials describe such artefacts as "wiggly worms", or "connected structures", and they are an indication that the deconvolution strength ought to be dialled back. In my opinion, that is the case here. If I were to process an image like this, I would do several trials with increasing (or decreasing) deconvolution strengths, and opt for the one with visually the best detail, without showing signs of "wiggly worms".

[SamAndrew]

Well I've now taken myself down a rabbit hole processing Hubble data. Tried BlurXterminator; what an improvement! 😂😅 feels like I'm seeing the univese in more resolution that anyone has ever done before 😂

[SamAndrew]

On 25/04/2023 at 18:26, ollypenrice said:

The main outer glow may also be a bit blue but, hey, the image has something to say. lly

Interestingly the raw data with basic stretch applied has a blue halo. Photometric colour calibration unsurprisingly fails at this pixel scale, so who knows what the right colour is 🤷‍♂️

[powerlord]

23 hours ago, vlaiv said:

Could you post - just a crop of linear data of that region?

I'm intrigued to figure out what has happened with the data in order to display that web like feature. I have an idea, but would need the data to confirm it.

We see it as a web - and AI made in a web like structure because it is lacking color differentiation present in Hubble image. Out of 5 spikes going out from the center - top one is actually of the different color in HST image and represents background gap rather than some material. After being blurred - it start to have very similar shape to other features blurred (while features are different - blurred versions look morphologically the same) and since red channel is clipped - color also starts being the same (difference being in non clipped portion of the red channel - that is what I'm hoping to find in linear data).

Right so firstly - apologies there - because what I posted it turns out was NOT the raw stack. I'd been faffing with this so many times I had a directory full of 100s of version with contructive names like M82.230419_230415-RGB-session_1-crop-lpc-cbg, M82.230419_230415-RGB-session_1-crop-lpc-cbg-1, M82.230419_230415-RGB-session_1-crop-lpc-cbg-1b etc

I thought I'd selected the raw one, but it was not - it was one that had been through various experiments in deconvolution and wavelets in siril, various faffing about in affinity, noiseX, etc.

And yeh - I reckon during that it got 'webbed'.

I've attached the almost definitely (but can't guarantee it) original stacked, cropped and lpced image out of AAP (rgb only, no Ha) And when I experimented with deconvolving and waveletting that puppy, I got webs too if I went too far.

M82.230419_230415-RGB-session_1-crop-lpc-cbg.fitsAs I say, back to the drawing board for attempt number 1.4x10⁴e...

Edited April 27, 2023 by powerlord

[powerlord]

12 hours ago, SamAndrew said:

Interestingly the raw data with basic stretch applied has a blue halo. Photometric colour calibration unsurprisingly fails at this pixel scale, so who knows what the right colour is 🤷‍♂️

there u go, I knew it! I'm right, and NASA is wrong. I'll expect an apology from them forthwith. too much blue indeed... 🙄

 

[powerlord]

12 hours ago, SamAndrew said:

Well I've now taken myself down a rabbit hole processing Hubble data. Tried BlurXterminator; what an improvement! 😂😅 feels like I'm seeing the univese in more resolution that anyone has ever done before 😂

Do you reckon NASA is buying site licences from Russel to redo all Hubble's images ? Do you reckon they'd tell anyone if they did ?

😃

[vlaiv]

37 minutes ago, powerlord said:

I've attached the almost definitely (but can't guarantee it) original stacked, cropped and lpced image out of AAP (rgb only, no Ha)

I took a look and I'm not any wiser. I think that it is simply artifact of oversampling + sharpening.

[AstroMuni]

1 hour ago, powerlord said:

Do you reckon NASA is buying site licences from Russel to redo all Hubble's images ? Do you reckon they'd tell anyone if they did ?

😃

Now you are getting into areas like the ZWO ASIAir controversy. 😉 

[windjammer]

PL - I think if you painted out the kangaroo, jello and extra disciples then Vlaiv would be happy. He knows what he likes!

[wimvb]

@powerlord, I took the liberty of downloading the fits file you shared and processed two versions in PixInsight. The difference is that in one versioin I used BlurXTerminator, and in the other I used conventional deconvolution. No other sharpening processes were used, only stretching and colour saturation. I also refrained from using any masks in PixInsight to selectively do any processing. So no star reduction beyond what the deconvolution steps did.

In conventional deconvolution it's very easy to over do it and introduce artefacts, I tried to avoid that as much as possible, while still having noticable sharpening. It is obvious that BXT allows much stronger deconvolution than PixInsight's deconvolution process. It can sharpen the image much closer to the noise floor, and decreases stars much more. I used BXT with a strength of 0.70, and custom PSF size.

This image (saved as jpeg at highest quality) lacks the H-alpha, so not as smoking as your original.

[vlaiv]

3 minutes ago, wimvb said:

I used conventional deconvolution

When you say that - what exactly is deconvolution method used?

[wimvb]

51 minutes ago, vlaiv said:

When you say that - what exactly is deconvolution method used?

With point spread function extracted from non saturated stars in the central part of the image.

I hope this answers your question.

[vlaiv]

21 minutes ago, wimvb said:

I hope this answers your question.

Yep.

It's Richardson-Lucy or LR deconvolution (regularized).

I have suspicion that we are using LR deconvolution the wrong way

It is a very good algorithm - but it was developed to deal with single sub rather than with stack. Algorithm is using probabilistic approach and assumptions are that noise distribution follows Poisson + Gaussian pattern - meaning shot noise + some amount of read noise.

When applying it to the stack - we are using it outside of its intended domain. In ideal world where we stacked with shift + add technique - i.e. no sub pixel shifts of subs and no rotation of subs for alignment prior to stacking - it would be still valid, but since we use alignment with interpolation that effectively changes noise distribution - I wonder how efficient LR really is in that scenario.

Btw, this is just a technical rant on my part, nothing really to do with your comparison, I just wondered which deconvolution method you used.

[The Lazy Astronomer]

So what, if any, would be the most appropriate algorithm for deconvolution of stacked images?

[vlaiv]

1 minute ago, The Lazy Astronomer said:

So what, if any, would be the most appropriate algorithm for deconvolution of stacked images?

That is a good question, and I must admit that I have no clue.

At one point I did some research into different deconvolution algorithms and while we have "ideal" algorithm that produces accurate results in theoretical conditions - which is just inverse filtering - or division in frequency domain (as convolution is simply multiplication in frequency domain - and inverse is division) - problem is that in real conditions we have two issues with our input.

1. We don't have actual convolved function to perform deconvolution on. What we have is such function polluted / distorted by presence of noise.

2. We don't have knowledge of PSF used to convolve our baseline function

We can extract it from stars in the image, but it will only be approximation to a certain degree because in reality - we don't have single PSF operating on our image. We have a range of PSFs that are very similar - but not identical. Stars have different spectra and seeing influence depends on wavelength of light in question. Blue stars will almost certainly have higher FWHM than red ones - when imaged thru reflector. Using refractor adds another level of complexity as no refractor is ideally corrected.

Then there are aberrations of optical system that vary with position in field (coma for example - grows as function of distance from optical axis) and even with correctors - we don't get stars that are equal all over the frame.

Further - we have impact of seeing that is not the same over the whole FOV. With longer exposure it tends to average out - but still - it won't be 100% the same in all points on the image.

Given two points above - we can only hope to achieve so much with deconvolution. Different algorithms deal with above two points in different way. LR deconvolution assumes fixed blur kernel and Gaussian+Poisson distribution - which would be single exposure. More data resembles that - better results it will produce.

I don't remember seeing comprehensive comparison of deconvolution algorithms on astronomical data (specifically stacked amateur images).

[wimvb]

53 minutes ago, vlaiv said:

Btw, this is just a technical rant on my part, nothing really to do with your comparison, I just wondered which deconvolution method you used

I only apply deconvolution in areas that have a high SNR, ie bright parts. So the exact noise distribution should be of minor concern. While it is possible to use deconvolution without a mask, and use the regularisation parameters to control the process in low SNR areas, I take the easy way out and mask off low SNR areas.

[pipnina]

A lot of heavy technical discussion in this thread.

I'm happy to enjoy the great picture myself!

[AMcD]

Using “Blur Xterminator” on images of black holes 😂

 

[powerlord]

Rl is the recommended decov type in siril too.and seperate wavelet sharpening.

I keep pestering Russell to do a ps plugin (for affinity photo) or a standalone version of blurx, but his concern is he will get flooded with support requests by folk using it wrongly on on linear images. 😞