M57 + Arc Hyperbolic Sine Stretch

[Galen Gilmore]

I just processed a picture I took a couple months ago because I only just got around to doing it. But I also just happened to stumble across an interesting YouTube video about a processing technique called the Arc Hyperbolic Sine Stretch.

I decided on giving it a go on M57, and all I have to say is WOW!

The technique did a great job of making the image feel very natural by preserving star color, and preventing them from becoming bloated.

If anyone else is interested in it, here is the video.

 

The image is about 1h 22m of 2m exposures from my SW 130PDS, and Canon 600D from my light polluted front yard.

[JamesF]

On 18/10/2020 at 05:04, Galen Gilmore said:

I just processed a picture I took a couple months ago because I only just got around to doing it. But I also just happened to stumble across an interesting YouTube video about a processing technique called the Arc Hyperbolic Sine Stretch.

I decided on giving it a go on M57, and all I have to say is WOW!

The technique did a great job of making the image feel very natural by preserving star color, and preventing them from becoming bloated.

If anyone else is interested in it, here is the video.

There are some comments in that video that bother me...

The first is one about the green histogram being further to the right because there are twice as many green photosites on his colour sensor than there are red or blue.  I'm not sure I follow the logic of that.  Surely the green histogram would shift to the right if the green photosites counted more photons, or if more than one green pixel value were added into the demosaic process without any scaling of the values used (which I don't believe will happen).  I'd guess that the reason is actually that the green photosites are just more sensitive and count more photons?

The second is the moving of the black point in the individual channels before the stretch is done.  If the point of this particular stretch is to preserve the colour balance, then surely moving the black points beforehand defeats that goal because in doing so the colour balance is changed?

James

[Galen Gilmore]

3 hours ago, JamesF said:

There are some comments in that video that bother me...

The first is one about the green histogram being further to the right because there are twice as many green photosites on his colour sensor than there are red or blue.  I'm not sure I follow the logic of that.  Surely the green histogram would shift to the right if the green photosites counted more photons, or if more than one green pixel value were added into the demosaic process without any scaling of the values used (which I don't believe will happen).  I'd guess that the reason is actually that the green photosites are just more sensitive and count more photons?

The second is the moving of the black point in the individual channels before the stretch is done.  If the point of this particular stretch is to preserve the colour balance, then surely moving the black points beforehand defeats that goal because in doing so the colour balance is changed?

James

Im not knowledgable enough to really comment on everything that you mentioned, but isn't a green color cast after debayering normal on the ZWO asi294? 

And for the second point, I see what you mean, it does seem a bit odd. But surely it doesn't change color balance too much, considering you are eyeballing the colorbalance, and you are only changing the black point per individual channel by only 1 - 2 positions.

[JamesF]

41 minutes ago, Galen Gilmore said:

isn't a green color cast after debayering normal on the ZWO asi294?

No idea, to be honest.  But if it is then my first guess would be that the green photosites are more sensitive or have a wider bandpass than the red or blue ones.

James

[Galen Gilmore]

23 minutes ago, JamesF said:

No idea, to be honest.  But if it is then my first guess would be that the green photosites are more sensitive or have a wider bandpass than the red or blue ones.

James

So I talked to the guy who made the video, this is what he had to say about the inquiries:

Quote

First, denosaicing is done for me using DSS, where an rggb debayer process is used and always results after stacking with the green channel having a higher SNR. That is because of having two green photosites as opposed to one. This is a very common result in OSC AP.

Secondly after adjusting the red and blue channels to align with the green channel, what occurs is a large offset in the black point with no data there and thus the black point needs to be “set” back to the left and remove the offset. This doesnt affect the color balance as there is no data being changed, it is just an offset adjustment of the black point. I listed a link in my video to Mark Shelly’s website, he came up with this technique and he is the author of the Pixinsight arcsinh stretch module. It requires no offset in the black point, please read his explanation to understand it properly.

Hope this helps clarify. Also, “the proof is in the eating of the pudding”.

Cheers and thanks for watching,

AstroEd

 So take from that what you will. Again I'm not very experienced when it comes to all of the technical details, all I know is that the process produces some very nice results from what I have seen so far. But there is no perfect process for every situation, what is described in the video can probably be improved upon in one way or another, but I'm not the person to ask when it comes to that. 😅

[JamesF]

I don't know.  It just doesn't feel right as an explanation to me.  I'm not sure if DSS stacks and then demosaics, or demosaics and then stacks.  My immediate thought is that it would be better to stack the raw frames as generating the full colour image acts to "smooth out" the data, but I'm not certain.  I don't get the bit about SNR either.  I'd have to work through the maths to be certain.  Perhaps there's a dependency on the stacking method used.

I don't doubt that Mark's arcsinh stretch works, mind.  I've used that myself and been happy with the results.  The way he implemented it in PS is clever too.  I'm not sure I'd have thought of that myself.

James

[ollypenrice]

13 hours ago, JamesF said:

I don't know.  It just doesn't feel right as an explanation to me.  I'm not sure if DSS stacks and then demosaics, or demosaics and then stacks.  My immediate thought is that it would be better to stack the raw frames as generating the full colour image acts to "smooth out" the data, but I'm not certain.  I don't get the bit about SNR either.  I'd have to work through the maths to be certain.  Perhaps there's a dependency on the stacking method used.

I don't doubt that Mark's arcsinh stretch works, mind.  I've used that myself and been happy with the results.  The way he implemented it in PS is clever too.  I'm not sure I'd have thought of that myself.

James

In my understanding it has to demosaic first because colour must be attributed according to the Bayer pattern of that particular chip. If it aligned on the stars first the alignment would ignore the Bayer matrix and would stack a red pixel onto a blue onto a green, etc etc, and the colour information would be lost.

I'll need to watch the video again but I'm left shaking my head when I hear that there is only one best way to stretch data. The stretch is the most important intervention in processing and cannot be reduced to a single method for three reasons. 1) Every image is different. 2) An imager might have a specific set of objectives for a dataset. Sometimes this can be extracted at the expense of that. We prioritize what it is we want our image to show. 3) Some of the most effective stretching techniques work by blending different stretches simply because one stretch cannot do everything.  This doesn't mean I don't think Mark's stretch is an excellent one. It is.

Olly

[JamesF]

35 minutes ago, ollypenrice said:

In my understanding it has to demosaic first because colour must be attributed according to the Bayer pattern of that particular chip. If it aligned on the stars first the alignment would ignore the Bayer matrix and would stack a red pixel onto a blue onto a green, etc etc, and the colour information would be lost.

A good point.  I'm clearly out of practice with all this OSC stuff :)

James

[Galen Gilmore]

On 22/10/2020 at 03:06, ollypenrice said:

In my understanding it has to demosaic first because colour must be attributed according to the Bayer pattern of that particular chip. If it aligned on the stars first the alignment would ignore the Bayer matrix and would stack a red pixel onto a blue onto a green, etc etc, and the colour information would be lost.

I'll need to watch the video again but I'm left shaking my head when I hear that there is only one best way to stretch data. The stretch is the most important intervention in processing and cannot be reduced to a single method for three reasons. 1) Every image is different. 2) An imager might have a specific set of objectives for a dataset. Sometimes this can be extracted at the expense of that. We prioritize what it is we want our image to show. 3) Some of the most effective stretching techniques work by blending different stretches simply because one stretch cannot do everything.  This doesn't mean I don't think Mark's stretch is an excellent one. It is.

Olly

I suppose you have to oversell it a bit in order to get clicks on Youtube. You guys seem rather familiar with Mark's Stretch, where as this video is my first introduction to it.

Also, I would always trust the word of my fellow SGL members over a video I found on Youtube, haha.

[Seelive]

Asinh stretching preserves the RGB ratio after the stretch whereas 'normal' curve stretching doesn't. It's fairly easy to show using the formulas involved for asinh and 'normal' curve stretching. It's a technique that's been around for years. Even my copy of IRIS dating from 2010 implements it.