Pixel scale

[StuartT]

So I have gathered that the ideal pixel scale is 1-2 arcsec per pixel (depending on seeing). But I have seen very nice images significantly more oversampled than this rule.

For example, this image from astrobin https://www.astrobin.com/full/p2adjl/0/ is at 0.658 arcsec/px and looks good to me.  

My setup is at 0.958 and I was a little concerned I was oversampling, but maybe I don't need to worry.

View on the 1-2 arcsec/px rule? When is it ok to break it?

Thanks

[vlaiv]

It is a tradeoff.

If you oversample image - there is nothing going to happen in terms of it looking bad or anything. When you view it in "fit to screen" - it will look as properly sampled image because software will scale it down so it fits inside your window.

Problem with oversampled image comes when you look at it at 100% - as far as visual part goes. Look at the image you linked to at 100% zoom:

This is just a crop of it. Stars look large - but that is not the main issue - in above crop that large "blob" star near center - is actually two stars that are not resolved.

Look at that image sampled at 2"/px (which is much closer to actual resolution), when viewed at 100%:

Ok, now sampling rate matches level of detail.

If you want to see what image of star field looks like when properly sampled at 0.658"/px - then

I scaled right image to match resolution and that is closer to actual pixel scale. Stars are simply tighter and detail is there that matches resolution. Yes, left image does not go as deep - but it simply does not have detail of right image.

(it has also been deconvolved in order to try to recover some resolution - which caused SNR loss).

In the end - when you oversample you loose SNR in comparison to properly sampling. That is the main problem. In order to get noise free image that goes deep - you need to spend much more time imaging than if you properly sampled.

[StuartT]

2 hours ago, vlaiv said:

It is a tradeoff.

If you oversample image - there is nothing going to happen in terms of it looking bad or anything. When you view it in "fit to screen" - it will look as properly sampled image because software will scale it down so it fits inside your window.

Problem with oversampled image comes when you look at it at 100% - as far as visual part goes. Look at the image you linked to at 100% zoom:

 

This is just a crop of it. Stars look large - but that is not the main issue - in above crop that large "blob" star near center - is actually two stars that are not resolved.

Look at that image sampled at 2"/px (which is much closer to actual resolution), when viewed at 100%:

 

Ok, now sampling rate matches level of detail.

If you want to see what image of star field looks like when properly sampled at 0.658"/px - then

 

I scaled right image to match resolution and that is closer to actual pixel scale. Stars are simply tighter and detail is there that matches resolution. Yes, left image does not go as deep - but it simply does not have detail of right image.

(it has also been deconvolved in order to try to recover some resolution - which caused SNR loss).

In the end - when you oversample you loose SNR in comparison to properly sampling. That is the main problem. In order to get noise free image that goes deep - you need to spend much more time imaging than if you properly sampled.

Thanks vlaiv. I can always rely on you for a detailed answer.

So would you say my current setup is ok (0.958)? Which side of the 1-2 is it best to stray on (oversampled or undersampled)? Because we only have the cameras and scopes we have, so we have limited ability to change pixel scale unless we buy a new expensive piece of equipment.

Edited September 12, 2021 by StuartT

[vlaiv]

15 minutes ago, StuartT said:

So would you say my current setup is ok (0.958)? Which side of the 1-2 is it best to stray on (oversampled or undersampled)?

That really depends on scope in question and sky conditions.

In general - I don't think that 1"/px is feasible resolution for most people. 8"+ of aperture, premium mount and those few nights a year when seeing is great - then yes, 1"/px could be pulled off.

I'd say that for most people - it would be somewhere in the middle - 1.5-1.8"/px. If you use small scope - like 4" or less - just go for 2"/px without worrying too much about it.

Resolution of image to some degree depends on scope aperture as well (it sort of goes into the mix).

If you want to know what sort of resolution your setup makes - just take some of your images - stacked data in linear stage before processing and look at FWHM of stars in that image. Divide value with 1.6 - that is resolution you should be aiming for.

If you get that your FWHM is closer to 1.6" - then go for 1"/px, but if it is 3" or more - then 2"/px makes more sense.

[The Lazy Astronomer]

49 minutes ago, StuartT said:

Thanks vlaiv. I can always rely on you for a detailed answer.

So would you say my current setup is ok (0.958)? Which side of the 1-2 is it best to stray on (oversampled or undersampled)? Because we only have the cameras and scopes we have, so we have limited ability to change pixel scale unless we buy a new expensive piece of equipment.

If you're oversampled (like likely you are), then all you need to do is bin the image - either during capture or post processing.

[StuartT]

11 hours ago, vlaiv said:

If you want to know what sort of resolution your setup makes - just take some of your images - stacked data in linear stage before processing and look at FWHM of stars in that image. Divide value with 1.6 - that is resolution you should be aiming for.

How do I measure FWHM? I assume you mean the FITS image that comes out of Astro Pixel Processor?

10 hours ago, The Lazy Astronomer said:

If you're oversampled (like likely you are), then all you need to do is bin the image - either during capture or post processing.

ok, this is a simple solution! So I can set binning in NINA to 2x2 I think. Does this mean I can increase exposure time and still have round stars?

[vlaiv]

9 minutes ago, StuartT said:

How do I measure FWHM? I assume you mean the FITS image that comes out of Astro Pixel Processor?

Yes, FITS from APP is linear data.

Not sure what you use for processing, but I've heard that PixInsight has FWHM measurement. If you don't have any software that will do it - look up AstroImageJ. It has star measurement tool and it will provide you with that info. Just be careful if you are measuring in pixels or in arc seconds (convert between them using your "/px ratio of original file).

Shift click will measure star that you click on and will show you FWHM in measurements table.

15 minutes ago, StuartT said:

ok, this is a simple solution! So I can set binning in NINA to 2x2 I think. Does this mean I can increase exposure time and still have round stars?

Round stars are best achieved by mechanically tuning your setup. If your guiding is good - they should be round and tight regardless of exposure length.

In principle, working on smaller resolution will create stars that appear tighter / rounder - due to different pixel scale, so yes, you should be able to expose for longer when binning - but again, that is not solving star shape issue - that is only masking it. Try to solve it properly - do it does not matter if you expose for 2 minutes or 5 minutes or 10 minutes - you should always get round stars (and as tight as possible).

[The Lazy Astronomer]

1 hour ago, StuartT said:

How do I measure FWHM? I assume you mean the FITS image that comes out of Astro Pixel Processor?

ok, this is a simple solution! So I can set binning in NINA to 2x2 I think. Does this mean I can increase exposure time and still have round stars?

DSS can give you a FWHM figure if you load your stacked image into it and analyse (not necessarily sure l particularly trust it to be accurate though - I've seen reports of it always giving a higher figure than PI).

As vlaiv said, good guiding is key to good star shapes. Ideally you'd be looking at wanting to get the RMS as reported by phd2 to be about half (or less!) of your image scale.

I think we have the same mount (EQ6-R?); l typically see guiding RMS of 0.6 - 0.8", which works well for my imaging scale of 1.7"/px. I've gone up to 5 minute subs (with narrowband filters) with no issue (theoretically no need for me to expose for any longer than that).

Your scope's a bit bigger and beefier than mine, but if you see similar RMS, then binning 2x2 should "fix" any star shape issues you experience on longer exposures. 

[StuartT]

7 hours ago, The Lazy Astronomer said:

DSS can give you a FWHM figure if you load your stacked image into it and analyse (not necessarily sure l particularly trust it to be accurate though - I've seen reports of it always giving a higher figure than PI).

As vlaiv said, good guiding is key to good star shapes. Ideally you'd be looking at wanting to get the RMS as reported by phd2 to be about half (or less!) of your image scale.

I think we have the same mount (EQ6-R?); l typically see guiding RMS of 0.6 - 0.8", which works well for my imaging scale of 1.7"/px. I've gone up to 5 minute subs (with narrowband filters) with no issue (theoretically no need for me to expose for any longer than that).

Your scope's a bit bigger and beefier than mine, but if you see similar RMS, then binning 2x2 should "fix" any star shape issues you experience on longer exposures. 

Ok, I should probably point out that

a) I am using a 150mm apo with reducer (focal length 808mm)

b) I am not guiding.

But given that my focal length and camera are both fixed, there is not much I can do about pixel scale anyway, right?

[vlaiv]

13 minutes ago, StuartT said:

b) I am not guiding.

I'd put this high on your priority list - get guiding running on your imaging rig.

14 minutes ago, StuartT said:

But given that my focal length and camera are both fixed, there is not much I can do about pixel scale anyway, right?

Pixel binning has the same effect as changing pixel size. You take group of 2x2 or 3x3 (or higher number) of pixels and treat them as single pixel by simply adding the light in each - same would happen if we had larger pixel to begin with - it would capture all light falling on it.

If you change "pixel size" - you effectively change pixel scale although your focal length remains the same.

[ollypenrice]

I very much like imaging at a pixel scale similar to yours, in my case about 0.9"PP. I find I can present images at full size and be happy with how they look. Ironically, perhaps, I find I use this high res setup to do single panel images which I often then crop to post areas of interest at full size. I also run a widefield rig at much lower resolution (3.5"PP) and I usually make mosaics with that one. So my small field images get smaller and my widefields get wider...  It just seems to work out that way.  Be aware, though, that aiming to post at full size means taking plenty of exposures and you need a guide RMS no more than half your image scale.

0.9"PP gives me results like this:

https://www.astrobin.com/full/miqpyu/0/?mod=&real=

https://www.astrobin.com/full/335042/0/

https://www.astrobin.com/full/393219/0/

https://www.astrobin.com/full/419975/0/

I've imaged at higher resolution but not found that it gave better results so I stick at 0.9 by personal preference.

Olly

 

[vlaiv]

I'll use one of Olly's images to demonstrate that what looks perfectly fine at 0.9"/px - in fact has less detail than that and can be happily sampled at say ~1.5"/px instead.

Here is crop of original image:

I took that crop and resampled it to 2/3 of original size - which is 1.35"/px if original was at 0.9"/px. Here is what that looks like:

Now, if original version has some details that can't be recorded at 1.35"/px - then if I enlarge this smaller version - we should be able to tell that it is different than original crop above. So I did exactly that, here is small version enlarged back to 0.9"/px:

Can we spot any differences between these two (and no, it is not the same image - I really did reduce the size and enlarged back again )?

[StuartT]

56 minutes ago, vlaiv said:

I'd put this high on your priority list - get guiding running on your imaging rig.

I avoid guiding by keeping my subs short (35s). This gives me round stars with my setup. I read somewhere that guiding used to be more crucial with CCD cameras because you needed a longer exposure to swamp the read noise. But with newer, low read noise cameras like mine (ASI 2600MC) this is much less significant, so there is little advantage in doing smaller numbers of long exposures compared to larger numbers of short exposures. Plus, it saves all the hassle of setting up guiding! 

55 minutes ago, vlaiv said:

Pixel binning has the same effect as changing pixel size. You take group of 2x2 or 3x3 (or higher number) of pixels and treat them as single pixel by simply adding the light in each - same would happen if we had larger pixel to begin with - it would capture all light falling on it.

If you change "pixel size" - you effectively change pixel scale although your focal length remains the same.

So I'm confused. Earlier in this thread, you seemed to be saying that binning was not a good strategy? Or maybe I am misunderstanding you.

 

[AKB]

26 minutes ago, vlaiv said:

we should be able to tell that it is different than original crop above. So I did exactly that, here is small version enlarged back

Of course, visual inspection is not a particularly good test.  (In the weekend papers they have a children's puzzle section with "Spot the difference": 6 differences, ... I'm lucky to see four of them!)

So I took both those images, subtracted them (adding a small offset) and multipled the result by 30...

...OK, it is correlated with the image, but I think we'd have to agree that the difference is really "in the noise".

Tony

[vlaiv]

2 minutes ago, StuartT said:

I avoid guiding by keeping my subs short (35s). This gives me round stars with my setup. I read somewhere that guiding used to be more crucial with CCD cameras because you needed a longer exposure to swamp the read noise. But with newer, low read noise cameras like mine (ASI 2600MC) this is much less significant, so there is little advantage in doing smaller numbers of long exposures compared to larger numbers of short exposures. Plus, it saves all the hassle of setting up guiding!

Did you actually measure level of read noise compared to other noise sources?

Most dominant noise source is light pollution noise. That however varies with location, but also with sampling rate.

You might be located in heavy LP - but similarly as target signal - LP also is "a signal" - unwanted one that brings its own noise into mix, but it still acts as signal. If you over sample when capturing you are in effect reducing LP signal in the same way you reduce target signal thus not gaining anything like lowering effective light pollution - but you do reduce level of it and hence noise.

This means that LP noise becomes smaller - and you need to spend more time on single frame to swamp read noise.

For example - if you image at 1"/px versus 2"/px - you need four time as long exposure to equally swamp read noise with LP noise. Binning does not help here - as it happens after you read out the pixels and read noise is already added to the mix. It helps with SNR as is after read noise has been added.

7 minutes ago, StuartT said:

So I'm confused. Earlier in this thread, you seemed to be saying that binning was not a good strategy? Or maybe I am misunderstanding you.

Now I'm confused - I was under impression that I always advocated for proper sampling rate and in cases where this means binning - to go with binning, unless you are able to actually exploit resolution you are working at.

[vlaiv]

3 minutes ago, AKB said:

So I took both those images, subtracted them (adding a small offset) and multipled the result by 30...

This is actually the proper way to asses differences - but I cheated a bit here. I already knew that there won't be perceivable difference as such test needs to be performed on linear scale - and often differences are order of 1/1000 of pixel values. In above image visible differences are more down to 8bit samples being compared than anything else. Data is also stretched which changes some of frequency response - in linear regime difference is even smaller.

In fact - when I asses what is proper sampling rate - I go by SNR - what is level of detail that can be sharpened in image without making noise too visible.

[ollypenrice]

17 hours ago, vlaiv said:

I'll use one of Olly's images to demonstrate that what looks perfectly fine at 0.9"/px - in fact has less detail than that and can be happily sampled at say ~1.5"/px instead.

Here is crop of original image:

I took that crop and resampled it to 2/3 of original size - which is 1.35"/px if original was at 0.9"/px. Here is what that looks like:

Now, if original version has some details that can't be recorded at 1.35"/px - then if I enlarge this smaller version - we should be able to tell that it is different than original crop above. So I did exactly that, here is small version enlarged back to 0.9"/px:

Can we spot any differences between these two (and no, it is not the same image - I really did reduce the size and enlarged back again )?

Yes, that's a good test! I need bigger pixels. 

Olly

[StuartT]

Ok, vlaiv. I clearly misunderstood you. I was reading this remark as negative about binning: 

Quote

you should be able to expose for longer when binning - but again, that is not solving star shape issue - that is only masking it. Try to solve it properly 

But I understand now what you are saying (I think). Thanks for taking the time to explain.

 

21 hours ago, vlaiv said:

Did you actually measure level of read noise compared to other noise sources?

I did not. This is bit beyond my knowledge at the moment. I am still very new to all this. I think I was taking this from various other sources about CMOS cameras.

[vlaiv]

6 hours ago, StuartT said:

I did not. This is bit beyond my knowledge at the moment. I am still very new to all this. I think I was taking this from various other sources about CMOS cameras.

In principle that is right - CMOS sensors have lower read noise and hence require less exposure length, but one should really measure how much read noise there is compared to other noise sources.

I know that author of SharpCap did YouTube video on how and why regarding all of that - and I believe it is a good material to watch in order to understand.

[Lee_P]

On 12/09/2021 at 22:46, vlaiv said:

That really depends on scope in question and sky conditions.

In general - I don't think that 1"/px is feasible resolution for most people. 8"+ of aperture, premium mount and those few nights a year when seeing is great - then yes, 1"/px could be pulled off.

I'd say that for most people - it would be somewhere in the middle - 1.5-1.8"/px. If you use small scope - like 4" or less - just go for 2"/px without worrying too much about it.

Resolution of image to some degree depends on scope aperture as well (it sort of goes into the mix).

If you want to know what sort of resolution your setup makes - just take some of your images - stacked data in linear stage before processing and look at FWHM of stars in that image. Divide value with 1.6 - that is resolution you should be aiming for.

If you get that your FWHM is closer to 1.6" - then go for 1"/px, but if it is 3" or more - then 2"/px makes more sense.

I've just come across this post and wonder if I could ask for some clarification. Is my understanding here correct:

I'm using an Askar 130PHQ and a 2600MC camera. This gives the potential resolution of my system as 0.78"/px. However, atmospheric conditions and mount inaccuracies mean that in reality the resolution is lower. To calculate what would be optimal for my equipment and sky conditions, I can take the FWHM of an image fresh from integration and divide by 1.6. (2.94 / 1.6 = 1.84"/px). So, 0.78"/px is definitely oversampled. If I bin2, then the working resolution is 1.56"/px, which is close to 1.84"/px. And as a sense check, it fits the general rule of thumb that between 1 and 2 "/px is usually good working resolution. 

The same idea, using old Askar FRA400 and 2600MC data:

Potential resolution of 1.93"/px. FWHM of 2.24/1.6 = 1.39"/px. No need to bin.

 

[vlaiv]

33 minutes ago, Lee_P said:

I've just come across this post and wonder if I could ask for some clarification. Is my understanding here correct:

I'm using an Askar 130PHQ and a 2600MC camera. This gives the potential resolution of my system as 0.78"/px. However, atmospheric conditions and mount inaccuracies mean that in reality the resolution is lower. To calculate what would be optimal for my equipment and sky conditions, I can take the FWHM of an image fresh from integration and divide by 1.6. (2.94 / 1.6 = 1.84"/px). So, 0.78"/px is definitely oversampled. If I bin2, then the working resolution is 1.56"/px, which is close to 1.84"/px. And as a sense check, it fits the general rule of thumb that between 1 and 2 "/px is usually good working resolution. 

The same idea, using old Askar FRA400 and 2600MC data:

Potential resolution of 1.93"/px. FWHM of 2.24/1.6 = 1.39"/px. No need to bin.

 

Yes.

I'd add the following to make things more clear:

- sampling rate (rather than potential resolution) is what you have when you pair certain focal length with certain pixel size (or pixel spacing to be even more correct).

- potential resolution of the system depends on aperture size, optical figure (diffraction limited optics or not - spot diagram RMS), mount tracking performance (guide RMS) and seeing conditions. In most cases we "calculate" for diffraction limited optics, although in some cases one should really account for spot diagram RMS if it is too large.

- you want to achieve good match between the two above - and first is easy to calculate, but second is easy to measure. Don't just settle for one image / one session - measure across sessions to get the feel for average FWHM you will get from your system as each night will be different. I've also found discrepancy between FWHM measurements in software - different software report different figures for some reason. I tent to trust ImageJ/AstroImageJ for this measurement.

In ImageJ you can't measure FWHM directly like in AstroImageJ (which has nice shortcut for that - just alt+click on star of interest) - but you can plot profile of a star and then fit gaussian shape to that profile to calculate FWHM. Both methods give very accurate answers on simulated gaussian profiles and agree on results (for stars that are round and horizontal profile in ImageJ).

To reiterate - arc seconds per pixel that you get for certain focal length and certain pixel size is not directly related to potential resolution. Rather, think of it millimeter scale on your caliper. Machining precision of caliper (how precisely it can physically measure) - that is potential resolution, that is telescope aperture + seeing + mount. It serves you no good to have very fine micrometer scale if your caliper is loose and you can't physically measure precisely enough.

 

[Lee_P]

Thanks vlaiv, insightful as ever.

1 hour ago, vlaiv said:

To reiterate - arc seconds per pixel that you get for certain focal length and certain pixel size is not directly related to potential resolution. Rather, think of it millimeter scale on your caliper. Machining precision of caliper (how precisely it can physically measure) - that is potential resolution, that is telescope aperture + seeing + mount. It serves you no good to have very fine micrometer scale if your caliper is loose and you can't physically measure precisely enough.

I'm having trouble getting my head around this bit. When I said "potential resolution", I meant if everything were perfect -- my telescope was transported into space, tracking was spot-on, and the optics were flawless. Would I then be able to achieve 0.78"/px?

[vlaiv]

21 minutes ago, Lee_P said:

I'm having trouble getting my head around this bit. When I said "potential resolution", I meant if everything were perfect -- my telescope was transported into space, tracking was spot-on, and the optics were flawless. Would I then be able to achieve 0.78"/px?

With said telescope - yes, but not with all telescopes.

How did you get that figure? I'm guessing that you took 1000mm of focal length of your scope and you took 3.76um pixel size and you calculated 206.3 * 3.76 / 1000mm = 0.76"/px, right?

Now if you take camera that has say 2.4um pixel size - you will get different sampling rate, but how is this related to what the telescope can resolve? You did nothing to the telescope itself, you just used different camera. Telescope remained the same and thus can't have its potential resolution changed.

If you want to know what your telescope can resolve in outer space with no need for tracking and without impact of atmosphere (and under assumption that your optics is diffraction limited) - then you can use planetary sampling formula which gives you that you need to have F/ratio that is at most x5 pixel size. For your telescope that is F/7.7 - so ideal pixel size would be 7.7 / 5 = 1.54um and corresponding sampling rate would be 206.3 * 1.54 / 1000 = 0.3177 = ~0.32"/px

That is maximum potential resolution of your telescope alone (and that is for blue light at 400nm - for say Ha that is 656nm - that is going to be different - slightly or about x1.5 times lower).

In any case - optics has potential to deliver certain resolution of image. Think of that as analog image. In order to properly digitize it - you need to sample at certain intervals (use certain pixel size). Using too fine sampling rate (too small pixels) - is waste of SNR as you don't need that fine pixel scale to record the image as is, and using smaller pixels just makes them receive lower signal each (same amount of light is spread over more pixels so in turn each get less light - less signal, lower SNR).

 

 

[Lee_P]

56 minutes ago, vlaiv said:

How did you get that figure? I'm guessing that you took 1000mm of focal length of your scope and you took 3.76um pixel size and you calculated 206.3 * 3.76 / 1000mm = 0.76"/px, right?

👍

Ok, analog / digital is a good way for me to understand it. Converting analog light into digital signal to then process.

Is it accurate to say that sampling rate is a measure of a telescope / camera combination's ability to record detail? And then an image's FWHM indicates how much resolution you've actually recorded? You want the two to match, so to do that you divide FWHM by 1.6, then if necessary bin your data so the sampling rate is close?

[vlaiv]

6 minutes ago, Lee_P said:

Is it accurate to say that sampling rate is a measure of a telescope / camera combination's ability to record detail? And then an image's FWHM indicates how much resolution you've actually recorded? You want the two to match, so to do that you divide FWHM by 1.6, then if necessary bin your data so the sampling rate is close?

Well, it depends.

I think that best way to think about it would be this - imagine telescope / mount system working without digital camera - with analog photo film. It will perform the way it does regardless of what is put at focal plane. Telescope and sky and physics in general does not know what is at focal plane nor does it care. It will produce image at focal plane with certain level of detail. This image, again does not depend on sampling rate, on pixel size, or anything like that - as you don't even need to have camera with pixels - you can use film.

FWHM is then measure of this ability of telescope / mount system (together with atmosphere) that characterizes how sharp the resulting image is. Again - it does not depend (this is not entirely true, pixels being area sampling devices do impact FWHM to small degree - but that is very complex topic) on pixels / camera used FWHM of signal will be what it is at focal plane.

After we have established that there is some image at focal plane, it is what it is and the fact that we are using pixels won't change that image - we can then address what pixels do and their ability to record the image, and it is quite simple, it goes like this (I'm again simplifying things for purpose of this explanation but effects that I'm neglecting are rather small and would unnecessarily complicate things / explanation):

too large - right size - too small

Your pixels can be too large, just right and too small.

If pixels are too large - then you are under sampling

if pixels are just right ("Goldilocks pixels") - then you are sampling good

if pixels are too small - then you are over sampling

Under sampling is not a bad thing. It just means that you might not capture all the detail there is in the image (and by detail - think detecting two close stars as two stars or oval blob where you are not certain what it is - one or more features - that is the meaning of "to resolve" - root of resolution thing).

Optimum sampling means that you'll be able to resolve all there is to be resolved in already formed image - and you will use the largest possible pixels to do so.

Over sampling mans that you will again be able to record / to resolve all there is to be resolved in image that is already formed in focal plane - but you will do that with smaller pixels than needed. This is hurting your SNR as smaller pixels simply means that you split light over more "buckets" than you need to and each "bucket" gets less light for that reason. Less light = lower signal = lower SNR / noisier image.

In above sense - pixel size is ability to record detail, however, FWHM is independent of that  - FWHM is intrinsic property of image. You can measure it from recorded data - and if you are over sampled or correctly sampled - you will measure correct FWHM (with small caveat of pixel blur, but again - technical, complex detail) and if you are under sampled - then you will start loosing detail and your FWHM will be off in its measurement for small amount (in fact - amount of error depends on how much you under sample and in usual cases, it's not that big of a deal). Again, in that sense - measurement of that FWHM from image that you've recorded does provide you with information of how sharp the image at focal plane was (regardless of what we used to record it).

Once you measure FWHM - then you have idea what are Goldilocks pixels in above sense - you take FWHM divide that value with 1.6 and this gives you sampling rate you should be aiming for (take that into account and your focal length and /or any focal reducers to get wanted pixel size - then bin accordingly or replace camera if that makes more sense - or as last option - don't bother - if you can live with SNR loss due to over sampling).