Reducer Myth: On Point

[Demonperformer]

19 minutes ago, vlaiv said:

How much lower?

And is this, maybe, the crux of the issue. If the resolution is still such that one cannot tell the difference when looking at it on whatever equipment you use (screen, printed copy, etc) then is there any detriment to using the focal reducer (with its attendant benefits)? I do take your point about when you are taking accurate measurements, but I suspect that AP-ers on the forum who take photos for that purpose are probably in the minority. The rest of us just want pictures looking as good as we can get them. And if we can achieve that with a reducer, why not?

[Demonperformer]

Just took a closer look at the cropped images. Although all the stars look 'blocky', it seems to me that the two brightest stars look less 'square' in the brighter image (the one taken with the reducer) than in the other one. And aren't rounder stars in images better?

[vlaiv]

1 minute ago, Demonperformer said:

And is this, maybe, the crux of the issue. If the resolution is still such that one cannot tell the difference when looking at it on whatever equipment you use (screen, printed copy, etc) then is there any detriment to using the focal reducer (with its attendant benefits)? I do take your point about when you are taking accurate measurements, but I suspect that AP-ers on the forum who take photos for that purpose are probably in the minority. The rest of us just want pictures looking as good as we can get them. And if we can achieve that with a reducer, why not?

Totally agree with you as I mentioned previously. I do encourage people that want to produce nice, sharp looking and presentable image to use focal reducers, to bin their images to increase SNR, etc. I'm also not saying that such images don't have scientific value.

I was just objecting to calling something theoretically backed to be a myth, without presenting suitable evidence to contrary. I also wanted to point out that deeper understanding of theory involved does have its benefits to both people interested more in scientific side of things as well as to people enjoying this hobby on a more casual level.

 

[vlaiv]

1 minute ago, Demonperformer said:

Just took a closer look at the cropped images. Although all the stars look 'blocky', it seems to me that the two brightest stars look less 'square' in the brighter image (the one taken with the reducer) than in the other one. And aren't rounder stars in images better?

Here again I must call upon theoretical side of things to explain how blocky stars happen in the first place. Undersampling is said to provide sampling artifacts because of loss of resolution. This is partly true. There is another essential bit to it. I also many times heard people wondering why very low resolutions don't show blocky stars as they would expect. Undersampling while cutting of true higher frequencies has the consequence of introducing higher harmonics of captured lower frequencies. This is how 'blockiness' happens. This is also reason why there are different resampling algorithms - they are designed to filter out these artificial higher harmonics (this is why they are often referred as to filters rather than algorithms - bilinear filtering, bicubic filtering, ....).

But most important things is actual frequency composition of original sampled signal. Due to nature of imaging process, signal that has been sampled by sensor is already filtered - there is Airy disk of optics which cuts off high frequencies and attenuates lower frequencies, then there is seeing and guiding error - both acting as low pass filter. These attenuate high frequency components in the image. When image is then undersampled and viewed, those high frequency harmonics that are created by sampling process have different shape if image is already prefiltered. This is why you need to seriously undersample in relation to seeing and scope aperture to start seeing blocky stars.

If one of posted images shows more blocky stars, and they have roughly similar sampling rate (factor 0.7) - it is sign that seeing was better during acquisition of blocky one, and in general there is undersampling for given seeing and scope aperture.

[vlaiv]

Here are results, you are probably going to laugh at minuscule difference, but indeed it is there.

Version with more noise is without reducer (obviously), and try not to mistake more noise for sharpness - look at stars and detail in top of the nebula cropped / outline of bright parts.

Here is more analytical approach to difference between images:

If detail were the same in both images - only thing present in this image would be noise. Fact that stars are showing is not necessarily sign of lack of detail - it shows impact of seeing on star position and FWHM difference between two stacks.

What does show difference in detail is that you can spot top of the nebula easily in this "missing detail" image. This means that that particular region in this crop suffered most from detail loss.

[Rodd]

1 hour ago, vlaiv said:

Ok, I'll state my view on that:

Using focal reducer will let you shoot the same object, which when cropped after will provide the same FOV as image taken without focal reducer. It will also provide you with higher SNR for the same acquisition parameters (sub time, number of subs) and same equipment (camera) compared to imaging without focal reducer (just how much SNR gain there will be, depends on number of other factors). Only part that I disagree with is that you will end up with same or more detail in the case of focal reducer vs native.

I am assuming that focal reducer does not act as corrective element (field flattening, coma correction, etc are out of the scope of this statement) and that it purely reduces focal length. I'm also assuming (as stated above) that all other parameters are the same - same camera, same sky conditions, ...

There is border line case where both using focal reducer and native will provide the same level of detail and that is if with both focal lengths, reduced and native, sampling resolution is higher than critical sampling for a given scope. In all other cases using focal reducer will indeed result in lower resolution over native focal length. How much lower? It also depends on number of factors, like sampling rate on both focal lengths (reduction factor of focal reducer), aperture size and sky conditions.

Well in this case the aperture was the same--in fact everything was the same, sub duration, # of subs etc.  It certainly dos not appear that the native resolution is higher.  There is enough data and 30 min subs--on a fairly bright DSO.   Acording to the myth--this should not be the case, as you suggest.  But it seems to be here.  

Rodd

[Rodd]

1 hour ago, Demonperformer said:

And is this, maybe, the crux of the issue. If the resolution is still such that one cannot tell the difference when looking at it on whatever equipment you use (screen, printed copy, etc) then is there any detriment to using the focal reducer (with its attendant benefits)? I do take your point about when you are taking accurate measurements, but I suspect that AP-ers on the forum who take photos for that purpose are probably in the minority. The rest of us just want pictures looking as good as we can get them. And if we can achieve that with a reducer, why not?

But this should not be the case according to the myth....myth being used as a title as apposed to an attribute.  The resolution of the unreduced image SHOULD BE higher--and appreciably so considering the focal length difference (1000 vs 700)

Rodd

[Rodd]

1 hour ago, Demonperformer said:

Just took a closer look at the cropped images. Although all the stars look 'blocky', it seems to me that the two brightest stars look less 'square' in the brighter image (the one taken with the reducer) than in the other one. And aren't rounder stars in images better?

Again--if true...contrary to the myth and the reason for the post!

Rodd

[Rodd]

11 minutes ago, vlaiv said:

Here are results, you are probably going to laugh at minuscule difference, but indeed it is there.

Version with more noise is without reducer (obviously), and try not to mistake more noise for sharpness - look at stars and detail in top of the nebula cropped / outline of bright parts.

Here is more analytical approach to difference between images:

If detail were the same in both images - only thing present in this image would be noise. Fact that stars are showing is not necessarily sign of lack of detail - it shows impact of seeing on star position and FWHM difference between two stacks.

What does show difference in detail is that you can spot top of the nebula easily in this "missing detail" image. This means that that particular region in this crop suffered most from detail loss.

I think , if I am interpreting this correctly, that this is evidence of a moral victory indeed!

Rodd

[vlaiv]

35 minutes ago, Rodd said:

I think , if I am interpreting this correctly, that this is evidence of a moral victory indeed!

Rodd

Well, if it is any consolation, here is what x2 software binning of native image produces in terms of detail loss:

And it increases SNR by x2, much more than using focal reducer.

[knobby]

So ... Reducers ... Yay or nay ?

Difficult decision !

[Rodd]

22 minutes ago, vlaiv said:

Well, if it is any consolation, here is what x2 software binning of native image produces in terms of detail loss:

And it increases SNR by x2, much more than using focal reducer.

But the myth does not cover software binning as far as I know.

[Rodd]

15 minutes ago, knobby said:

So ... Reducers ... Yay or nay ?

Difficult decision !

Not if you need the FOV increase--which is why I usually use them.

Rodd

[vlaiv]

8 minutes ago, knobby said:

So ... Reducers ... Yay or nay ?

Difficult decision !

Reducers have big advantage: they extend FOV. So depending on target, for casual imaging, why not?

We haven't even touched another aspect of possible use of focal reducers: Since they increase SNR for given imaging time, is there benefit of using focal reducers and afterwards doing some sort of drizzle to recover any resolution lost due to undersampling, and under which conditions would it be a good/viable option ....

[vlaiv]

1 minute ago, Rodd said:

But the myth does not cover software binning as far as I know.

Well, I don't know why you insist on calling it a myth, but that was besides the point, I made that post just to show you that even bigger undersampling has smaller impact on resolution / captured detail than one is lead to believe if we just examine how image looks. It really depends on the contents of the actual image. If image is already lacking high frequency components - cutting off those with undersampling will not have strong effect on image perception.

Using focal reducer is equivalent to binning in terms of detail lost. Both do the same thing (if you crop increased FOV from focal reducer) - alter sampling rate to lower than native. Side effect of both approaches is increase of SNR (trading resolution for SNR). If binning x2 in some cases has minimal impact on resolution/details why expect that "binning" ~ x1.429 would have larger effect?

[wimvb]

6 minutes ago, vlaiv said:

Reducers have big advantage: they extend FOV. So depending on target, for casual imaging, why not?

We haven't even touched another aspect of possible use of focal reducers: Since they increase SNR for given imaging time, is there benefit of using focal reducers and afterwards doing some sort of drizzle to recover any resolution lost due to undersampling, and under which conditions would it be a good/viable option ....

Drizzle does necessarily increase noise. The redundancy in the subs, that could be used for noise reduction, is now used to recover resolution. I believe that Margareth Thatcher once put ut so very well: there's no such thing as a free lunch.

[Rodd]

4 minutes ago, vlaiv said:

Well, I don't know why you insist on calling it a myth, but that was besides the point, I made that post just to show you that even bigger undersampling has smaller impact on resolution / captured detail than one is lead to believe if we just examine how image looks. It really depends on the contents of the actual image. If image is already lacking high frequency components - cutting off those with undersampling will not have strong effect on image perception.

Using focal reducer is equivalent to binning in terms of detail lost. Both do the same thing (if you crop increased FOV from focal reducer) - alter sampling rate to lower than native. Side effect of both approaches is increase of SNR (trading resolution for SNR). If binning x2 in some cases has minimal impact on resolution/details why expect that "binning" ~ x1.429 would have larger effect?

I call it the myth because that is its name.  You should ask Olly about it--I did not name it.  The myth came around because people believed that if you use a focal reducer on a galaxy say, to increase SNR you collect more data in a shorter period of time.  Then you can crop the image and enlarge and you will have the equivalent image captured faster.  The myth says this is not true, that the cropped and enlarged image will be of lesser resolution--quite a bit less.  But I think I have shown that the difference in resolution between the unreduced and reduced is so slight that it is, in fact, true enough.  that is the moral victory.

[vlaiv]

1 minute ago, wimvb said:

Drizzle does necessarily increase noise. The redundancy in the subs, that could be used for noise reduction, is now used to recover resolution. I believe that Margareth Thatcher once put ut so very well: there's no such thing as a free lunch.

Yes, I'm aware of that, but I just wonder what is the ratio of gain / loss. Of the top of my head I would say that you probably increase noise more with drizzle than you decrease with binning. But, there is important point that I can't account easily without analysis. Drizzle would be opposite of software binning while using focal reduce is more like using hardware binning - you cram more photons into single pixel, so while you add signal you still have one "dose" of read noise per increased signal unlike software binning where you have more.

[vlaiv]

1 minute ago, Rodd said:

I call it the myth because that is its name.  You should ask Olly about it--I did not name it.  The myth came around because people believed that if you use a focal reducer on a galaxy say, to increase SNR you collect more data in a shorter period of time.  Then you can crop the image and enlarge and you will have the equivalent image captured faster.  The myth says this is not true, that the cropped and enlarged image will be of lesser resolution--quite a bit less.  But I think I have shown that the difference in resolution between the unreduced and reduced is so slight that it is, in fact, true enough.  that is the moral victory.

Oh that myth, I don't believe it has anything to do with resolution/detail - it had to do with whether there is increase in SNR depending if target fits or does not fit on sensor, or something like that.

That one has been resolved long time ago ... I can't even remember what the myth was originally but I don't believe it had anything to do with resolution.

[Rodd]

4 minutes ago, vlaiv said:

Oh that myth, I don't believe it has anything to do with resolution/detail - it had to do with whether there is increase in SNR depending if target fits or does not fit on sensor, or something like that.

That one has been resolved long time ago ... I can't even remember what the myth was originally but I don't believe it had anything to do with resolution.

NO NO!  The SNR in the above cropped and enlarged sub IS WAY more than the native resolution image that was cropped and enlarged....that is absolutely contrary to the myth!  We need Olly!

[ollypenrice]

7 minutes ago, knobby said:

So ... Reducers ... Yay or nay ?

Difficult decision !

Not at all! Easy decision (ish!!!). If you are imaging a target which will fit on your chip with or without reducer you know certain things for sure. 1) The reducer will bring in no more photons from the object. Photons from the object are excluded or included only by the size of the objective. This is never debated by anybody, on either side of the debate. In the slightly belligerent modern parlance, which I don't like, it's a 'no brainer.' 2) The reducer will concentrate more object photons onto fewer pixels, 'filling' them faster. Nobody disagrees with this either. Where the debate lies is whether or not this really brings a useful gain since downsizing the unreduced image to the size of the reduced one will have more or less the same result.

So...reducers, Yay or nay? You are imaging M33 which will fit on your chip with or without reducer, let us say. Your aperture is fixed so, obviously, so is your photon count from M33. Do you put your fixed count of M33 photons onto more or fewer pixels? Choose fewer, with reducer, and you loose resolution but gain in SNR. Choose more, without reducer, and and you gain in resolution at the expense of SNR. But what happens if you downsize the unreduced image to the size of the reduced? I think the SNR becomes insignificantly different from the reduced. The great and the good will aaargue the maths this way and that but I think that, in a nutshell, it doesn't matter.

Olly

 

[Rodd]

3 minutes ago, ollypenrice said:

Not at all! Easy decision (ish!!!). If you are imaging a target which will fit on your chip with or without reducer you know certain things for sure. 1) The reducer will bring in no more photons from the object. Photons from the object are excluded or included only by the size of the objective. This is never debated by anybody, on either side of the debate. In the slightly belligerent modern parlance, which I don't like, it's a 'no brainer.' 2) The reducer will concentrate more object photons onto fewer pixels, 'filling' them faster. Nobody disagrees with this either. Where the debate lies is whether or not this really brings a useful gain since downsizing the unreduced image to the size of the reduced one will have more or less the same result.

So...reducers, Yay or nay? You are imaging M33 which will fit on your chip with or without reducer, let us say. Your aperture is fixed so, obviously, so is your photon count from M33. Do you put your fixed count of M33 photons onto more or fewer pixels? Choose fewer, with reducer, and you loose resolution but gain in SNR. Choose more, without reducer, and and you gain in resolution at the expense of SNR. But what happens if you downsize the unreduced image to the size of the reduced? I think the SNR becomes insignificantly different from the reduced. The great and the good will aaargue the maths this way and that but I think that, in a nutshell, it doesn't matter.

Olly

 

But what if you upsize the reduced?  The above reduced crop has a higher SNR than the original, Olly....tell me it doesn't  Why would I down size the original--I would upsize the reduced--that is the point

[vlaiv]

2 minutes ago, ollypenrice said:

Not at all! Easy decision (ish!!!). If you are imaging a target which will fit on your chip with or without reducer you know certain things for sure. 1) The reducer will bring in no more photons from the object. Photons from the object are excluded or included only by the size of the objective. This is never debated by anybody, on either side of the debate. In the slightly belligerent modern parlance, which I don't like, it's a 'no brainer.' 2) The reducer will concentrate more object photons onto fewer pixels, 'filling' them faster. Nobody disagrees with this either. Where the debate lies is whether or not this really brings a useful gain since downsizing the unreduced image to the size of the reduced one will have more or less the same result.

So...reducers, Yay or nay? You are imaging M33 which will fit on your chip with or without reducer, let us say. Your aperture is fixed so, obviously, so is your photon count from M33. Do you put your fixed count of M33 photons onto more or fewer pixels? Choose fewer, with reducer, and you loose resolution but gain in SNR. Choose more, without reducer, and and you gain in resolution at the expense of SNR. But what happens if you downsize the unreduced image to the size of the reduced? I think the SNR becomes insignificantly different from the reduced. The great and the good will aaargue the maths this way and that but I think that, in a nutshell, it doesn't matter.

Olly

 

Ah, but that is easy one

Would you rather bin in hardware or software? Same thing with focal reducer vs downsizing original image. Focal reducer acts as hardware binning, downsizing as software binning. Hardware binning has advantage over software as long as read noise is appreciable component of the noise.

[vlaiv]

5 minutes ago, Rodd said:

But what if you upsize the reduced?  The above reduced crop has a higher SNR than the original, Olly....tell me it doesn't  Why would I down size the original--I would upsize the reduced--that is the point

Effect is that you've gained some SNR at expense of detail in the image (depending on actual image signal it can be less or more obvious). Same thing happens when you bin and then upscale the image. Using focal reducer is the same as hardware binning and it is better than software binning. There is another important thing: with binning you do it on integer level (there are algorithms for fractional binning, but they don't behave predictably as integer binning), while with focal reducer you can do "fractional" binning (of sorts). Both do in essence the same thing: change sampling resolution to "coarser" from original or undersampling (this is probably not correct word because it relates to maximum image detail) - thus loosing detail of the image.

[Rodd]

1 minute ago, vlaiv said:

Effect is that you've gained some SNR at expense of detail in the image (depending on actual image signal it can be less or more obvious). Same thing happens when you bin and then upscale the image. Using focal reducer is the same as hardware binning and it is better than software binning. There is another important thing: with binning you do it on integer level (there are algorithms for fractional binning, but they don't behave predictably as integer binning), while with focal reducer you can do "fractional" binning (of sorts). Both do in essence the same thing: change sampling resolution to "coarser" from original or undersampling (this is probably not correct word because it relates to maximum image detail) - thus loosing detail of the image.

You miss my point.  If the reduced crop has a higher SNR than the unreduced original.....that is contrary to the myth, which states that the SNR will be the same once the images are size equalized