Planetary imaging - what do I need to know.....?

[CraigT82]

1 minute ago, newbie alert said:

Or could be just the weather, which if it isn't in your favour there's not much you can do about it

Well for that particular image he rates his seeing as 7-9/10, so no, not the weather.

 

[vlaiv]

22 minutes ago, CraigT82 said:

Here is a comparison of recent images from Chris Go and a few northern Europe planetary imagers.

This is in fact very interesting comparison as we have 12", 14" and 16" scopes and images produced with them.

Images are all of the same size - so we can compare who is sampling properly and who is not.

Here it is - from left to right, 12", 14" (Damien's EdgeHD) and finally 16" - these are frequency spectrums - log version, maximally stretched.

Properly sampled image will have circle touching edges of image - so all frequencies from center to the edge are used. When image is over sampled - frequencies towards the edge will be missing - there won't be signal at those high frequencies and there will be "dark" region towards the edge.

We can clearly see following from above spectra:

- each larger aperture has a bit more resolving power than the last as each spectra is larger in extent of frequencies

- all images are over sampled by at least factor of x2

- Mewlon is best planetary scope / sharpest optics (or weather conditions were the best) - as image could be sharpened the best - most rapid transition from region of signal to region without signal (best defined circle)

- other two scopes have softer outer edge of the circle - meaning they were not sharpened as well as they could have been (or maybe noise won't allow it) - signaling either inferior optics or poorer conditions.

[CraigT82]

5 minutes ago, vlaiv said:

This is in fact very interesting comparison as we have 12", 14" and 16" scopes and images produced with them.

Images are all of the same size - so we can compare who is sampling properly and who is not.

Here it is - from left to right, 12", 14" (Damien's EdgeHD) and finally 16" - these are frequency spectrums - log version, maximally stretched.

Properly sampled image will have circle touching edges of image - so all frequencies from center to the edge are used. When image is over sampled - frequencies towards the edge will be missing - there won't be signal at those high frequencies and there will be "dark" region towards the edge.

We can clearly see following from above spectra:

- each larger aperture has a bit more resolving power than the last as each spectra is larger in extent of frequencies

- all images are over sampled by at least factor of x2

- Mewlon is best planetary scope / sharpest optics (or weather conditions were the best) - as image could be sharpened the best - most rapid transition from region of signal to region without signal (best defined circle)

- other two scopes have softer outer edge of the circle - meaning they were not sharpened as well as they could have been (or maybe noise won't allow it) - signaling either inferior optics or poorer conditions.

Very interesting, thanks for that. So no it’s not Mr Go’s oversampling then as they’re all at it! 

[vlaiv]

13 minutes ago, CraigT82 said:

Very interesting, thanks for that. So no it’s not Mr Go’s oversampling then as they’re all at it! 

You can see that they are over sampling by size of Jupiter in their images.

Just for fun - let's do some calculations to see what sort of size Jupiter can be if properly sampled.

Let's take 16" scope (other will be just multiple of aperture size) and say 3.75um pixel size. That requires F/15 to sample properly

406mm at F/15 = ~6100mm of FL.

At that FL, 3.75um pixel will be sampling at 3.75 * 206.3 / 6100 = ~0.127"/px

Jupiter's largest apparent size is about 50" so when closest and properly sampled by 16" scope - it will give disk size in pixels that is 50/0.127 = ~394px in diameter.

8" scope will produce only half of that so about 200px and 4" - only 100px across.

[vlaiv]

By the way, look what happens when I reduce size of that Mewlon 300 image to 50% (green channel) and do spectral analysis on it:

It is now almost perfectly sampled (actually, now it's just a tiny bit undersampled - I should not have reduced it to 50% - maybe 60% or so was actual optimum size - I don't know what F/ratio was used to produce image so don't know how much it should be reduced - I eyeballed it to 50% of original)

 

[newbie alert]

3 hours ago, CraigT82 said:

Well for that particular image he rates his seeing as 7-9/10, so no, not the weather.

 

What is it 7/10 or 9/10?  I've seen better images from him in 7/10 conditions and he has frequent 9/10 seeing.. but they have a rainy season, winds from unfavorable directions etc etc..  

Think this has diverted from what I originally said , gone off in all sorts of tangents...none of which is applicable to what's been said here.. just to remind you

So if he had a C14 would that calculation still ring true? Christopher go uses a 290 with a pixel size of 2.9 so should he be using his at f11.6 or f14.6... so that calculation is abit vague and not really relevant

I used Chris as and example to highlight that he uses a 2.9 um camera on a C14 and most of the time uses a x2 Barlow, so he's imaging at f22, so if he's not doing it correctly then he's been doing it wrong for a good number of years, my point is for the same pixel size camera then the same formula should be applicable for a c9.25, or a c8 to get to the same f ratio as stated...  We all know that a bigger scope, with longer reach can resolve finer details, so how can the same formula be applicable to all sizes based on all sizes of scopes... Surely the example images that you have shown wouldn't all be big aperture/ long focal length scopes if the same applies to all? 

[vlaiv]

15 minutes ago, newbie alert said:

So if he had a C14 would that calculation still ring true? Christopher go uses a 290 with a pixel size of 2.9 so should he be using his at f11.6 or f14.6... so that calculation is abit vague and not really relevant

I used Chris as and example to highlight that he uses a 2.9 um camera on a C14 and most of the time uses a x2 Barlow, so he's imaging at f22, so if he's not doing it correctly then he's been doing it wrong for a good number of years, my point is for the same pixel size camera then the same formula should be applicable for a c9.25, or a c8 to get to the same f ratio as stated...  We all know that a bigger scope, with longer reach can resolve finer details, so how can the same formula be applicable to all sizes based on all sizes of scopes... Surely the example images that you have shown wouldn't all be big aperture/ long focal length scopes if the same applies to all? 

Ok, here is very simple logic that you can follow to understand why it holds true for both small and large scope.

12" scope will resolve double that of 6" scope - as it is twice as big. This means that we need to "zoom in" twice more with large scope to fully exploit this, right? This in turn means that focal length of larger scope needs to be twice as big as focal length of small scope.

Say that 150mm (6") scope is using 2250mm FL (F/15) - this means that 300mm (12") scope will need to use double that - 4500mm of focal length to zoom in twice as much. But then we have 300/4500 = F/15

We have same F/ratio.

When F/ratio is fixed - "zoom" follows the aperture size (because focal length follows aperture size). This is why we calculate F/ratio base on pixel size for ANY aperture size - as it holds true for both small and large scope.

But why do we calculate F/ratio based on pixel size?

Well - again, it is simple thing. If image in focal plane is of a certain size in millimeters and we use larger pixels - we will use less pixels to record the image.

But given certain amount of detail - there is exact number of pixels that are needed to record that detail, so if pixels are larger - we need to make image in focal plane larger - we change F/ratio (make it bigger) - because it increases focal length and size of image in focal plane.

[CraigT82]

1 hour ago, newbie alert said:

What is it 7/10 or 9/10

Well, both. Seeing varies throughout a session and even within a single capture. 
 

As you say we’re veering off topic. 

[newbie alert]

Yes I understand how a Barlow works, but in real life terms to get to f15 on a c8 would require a 1.5x Barlow,  the same for the 9.25 and 11 but a x1.365 Barlow for a C14.. can't see how it be the same Barlow strength for anything under the c14.. most use powermates which start at x2 .. so f20 or f22 depending on scope.. so not near the f11.6 or f14.6 mentioned earlier on the first page

[vlaiv]

19 minutes ago, newbie alert said:

Yes I understand how a Barlow works, but in real life terms to get to f15 on a c8 would require a 1.5x Barlow,  the same for the 9.25 and 11 but a x1.365 Barlow for a C14.. can't see how it be the same Barlow strength for anything under the c14.. most use powermates which start at x2 .. so f20 or f22 depending on scope.. so not near the f11.6 or f14.6 mentioned earlier on the first page

I'm not following. What does barlow lens and the way it works have to do with anything here?

Fact that powermates start at x2 and people using them on SCTs - has nothing to do with proper sampling. You can't expect laws of physics to change because you don't have suitable telecentric lens for your SCT

In any case - you don't benefit by using powermate over barlow for planetary - and with barlow, you can control magnification factor by utilizing certain distance from sensor.

It is very easy to dial in required barlow amplification to reach wanted F/ratio.

[newbie alert]

Powermates are telecentric . The op is using a SCT.. thought we comment on what's on the thread rather than going off on tangents and insulting people's equipment...

[vlaiv]

2 minutes ago, newbie alert said:

Powermates are telecentric . The op is using a SCT.. thought we comment on what's on the thread rather than going off on tangents and insulting people's equipment...

Now I'm completely lost ...

I'm guessing you are answering to my post above?

52 minutes ago, vlaiv said:

Fact that powermates start at x2 and people using them on SCTs - has nothing to do with proper sampling.

Here I'm just stating the fact that just because powermates start at x2 and that people use them on SCTs (and there is no reason not to use them visually for example) - has nothing to do with proper sampling.

If you want to sample at optimum rate and you have powermate and SCT, then why not choose camera with larger pixels?

54 minutes ago, vlaiv said:

You can't expect laws of physics to change because you don't have suitable telecentric lens for your SCT

I hope this was not perceived as me insulting peoples equipment?

Again, I was not speaking to anyone in particular nor putting down people that own SCT nor SCTs for that matter. I simply emphasized that laws of physics are as they are and can't change and if one chooses to use F/20 or F/22 with say 2.9um camera size - well they will be over sampled. Comment about suitable telecentric lens was again to the fact that indeed - there is no telecentric lens that magnifies less than x2 in sale (that I'm aware of).

In the end - I simply offered a solution - don't use telecentric lens, use barlow instead. For this application it makes no difference optically (like for visual where telecentric does not extend eye relief or for Ha solar where etalon performance will depend on it) and barlow can be tuned to needed magnification.

Alternative for anyone wanting to keep their telecentric lens and get optimum sampling is to change imaging camera. For F/20 for example, I'd use IMX429 mono with 4.5um pixel size.

[neil phillips]

On 27/09/2022 at 16:05, CraigT82 said:

Very interesting, thanks for that. So no it’s not Mr Go’s oversampling then as they’re all at it! 

Lol Craig All at it he he. 

I think the question should be. Surely, they all have been trying different sampling. And not just blindly sticking to one focal length?  And if that is true? Then clearly, they believe they are getting better images by oversampling by a factor of two.  Otherwise, why would they do it? I have got better images on occasion by oversampling. But of course, it's hard to swear it's the focal length. Could be just easier getting more precise focus at a larger scale. Or seeing improved coinciding with higher focal length?

It's a shame they can't be asked. I know one thing next year I am going to try a lot of experimenting with the high elevation the UK is going to have. I already have plans on getting a bigger scope for next year. So, this is all very interesting.

[vlaiv]

4 hours ago, neil phillips said:

I think the question should be. Surely, they all have been trying different sampling. And not just blindly sticking to one focal length?  And if that is true? Then clearly, they believe they are getting better images by oversampling by a factor of two.  Otherwise, why would they do it?

Could it be that:

a) people associate larger image with more quality (selfish gene, wanting bigger / better/ more or comparing one self with Hubble images)

b) If person A is making such a large image, why can't I?  (competitiveness)

c) General public's expectations and trying to get likable image?

I'm not saying that any of them is doing it consciously - I just know that I found my myself on several occasions in struggle of what is expected vs "what is right" and that tells me that the struggle is real.

[vlaiv]

Actually here is one explanation of which I was not aware previously for why might someone over sample:

Use of ADC - apparently it does correct for dispersion but adds optical aberrations and aberrations are smaller in higher F/ratio.

 

[inFINNity Deck]

58 minutes ago, vlaiv said:

Could it be that:

a) people associate larger image with more quality (selfish gene, wanting bigger / better/ more or comparing one self with Hubble images)

b) If person A is making such a large image, why can't I?  (competitiveness)

c) General public's expectations and trying to get likable image?

I'm not saying that any of them is doing it consciously - I just know that I found my myself on several occasions in struggle of what is expected vs "what is right" and that tells me that the struggle is real.

Hi Vladimir,

this is quite an interesting thread!

Regarding your above question I tend to resize my planetary images by 200% after processing mainly as it is easier to see the details (especially because I am presbyopic 🤓) and a bit because of c). So here is my latest Jupiter with C11 @ f/20 using a TeleVue 2x PowerMate, ZWO ADC, ZWO LRGB filters and a ZWO ASI174MM (5.9 micron pixels). The animation is the size as captured:

I think this is about the best we can expect from a C11 EdgeHD, only even better seeing may improve things (marginally). The best frame of this set I then resized by 200% to make the details stand out a bit more:

Obviously this is severely oversampled now and by nowhere as detailed as the images shown above.

Now, the reason I show this is that I tried to do this Frequency analysis you showed us above using Fiji's plugin FFTJ (both with my own images and that of Chris Go), but have trouble arriving at the same results as you did, although it appears that Chris's image is oversampled by a factor 4. Could you please explain the steps you took?

What I did is as follows:

- downloaded the full 4-Jupiter image
- cropped it to 862 x 812 pixels to only contain Chris's image
- made it greyscale
- loaded it in Fiji
- Using the FFTJ plugin generated a forward transformation at double precision
- Generated a Show Frequency Spectrum (logarithmic) at Volume-Center
- stretched it by clicking Auto several times (until maximum is reached) in the Brightness/Contrast function:

 

But as you can see I get a very different result, but, being a novice in Fiji/ImageJ, I take it I must be doing something wrong. 🤔

Nicolàs

[vlaiv]

13 minutes ago, inFINNity Deck said:

But as you can see I get a very different result, but, being a novice in Fiji/ImageJ, I take it I must be doing something wrong.

I worked on all three images except Chris's, but I think that we have the same result except for level of stretch.

I'll now do the same on Chris's version and will post results.

Here are the steps:

1. Load image in ImageJ

2. Make 512x512 selection centered on Chris's image (512x512 is just for FFT - much quicker computation than arbitrary ROI - and there won't be difference in spectrum as far as intensity goes - just phase).

3. Image type -> RGB stack

4. Select green channel in stack and separate it as single image

5. Image / type convert to 32bit

6. FFTJ / single precision (there is no difference really for this application) / Spectrum log centered at image center

7. Manual stretch

If I don't stretch it completely, here is what I get:

So that is pretty much as what you get, but if I stretch it to the limits here is what I get:

Now, you are right - image is not over sampled by factor of x2 - it is more over sampled by factor of say 2.5-3, and it looks like it was very affected by seeing or was not sharpened properly.

To my eyes, this is domain where there is some information:

However, as you move from center - there is point where MTF of recording drops significantly. If we want to represent this as MTF graph, it would look like this:

Which is consistent with poor seeing superimposed on telescope optics.

[vlaiv]

Here is another interesting exercise that we can do - we have two 14" scopes on that image.

Given that Jupiter image is roughly the same size on the image (we can't be certain at what date each image was taken and whether Jupiter's apparent size in the sky was the same) - we can compare if both of them are over sampled by equal amount. Do maximally stretched circles match in size?

Here it is - two C14 compared in frequency domain:

[inFINNity Deck]

Hi Vladimir,

thanks for explaining, managed to reproduce your results using FFTJ on the green channel in a 512x512px crop:

And this is when that green layer is reduced to 25%:

Indeed the oversampling seems to be a bit less than 4x.

Nicolàs

 

[inFINNity Deck]

Hi Vladimir,

here are the frequency spectra of my recording:

At the left the original recording, at the right the 200% resized version. I think it shows that at f/20 my set-up is not far off the optimum sampling rate. The camera has 5.9 micron pixels. According to my approach this should require a 5.9 x 3 = f/17.7 scope, according to yours 5.9 x 4 = f/23.6. So my f/20 set-up seems to be somewhere in between the two methods of calculating the correct focal ratio.

Can you please comment on this?

Nicolàs

[vlaiv]

7 minutes ago, inFINNity Deck said:

I think it shows that at f/20 my set-up is not far off the optimum sampling rate.

Yes, I would expect so.

One of difficulties of using above frequency spectrum approach for determining if image was properly sampled is SNR level in the image and how much image was sharpened.

 We are trying to estimate telescope MTF based on result it produces - but there are several unknowns in that process.

1) Actual frequency response of the target. Luckily here we know what the target is and we are sure that there are higher frequencies present (smaller detail is readily visible in Hubble images, so we know that we have not reached cloud "smoothness" level yet - in fact it is far away)

2) MTF of telescope

3) Seeing induced blur after stacking. While we use lucky imaging to minimize impact of the seeing, there is still some residual seeing that is left and is averaged out in stacking.

4) Noise due to sampling

5) Sharpening effects.

In the end profile in frequency domain might look like above graph. Black line is MTF of telescope, Orange line is what is restored in terms of image, but there is also orange jagged horizontal line that represents noise in the image. Noise is usually equally distributed in frequency domain and at some point it will cross signal line. When this happens - we have no way of determining where our signal line actually finishes.

Best case scenario is this (high SNR)

So noise is low compared to signal  and although we are wrong - we are wrong by "few pixels". We have a good estimate where signal finishes in frequency domain.

In above image vertical line would be representing our sampling points. Left of where MTF hits the zero would be under sampling, right of that point would be over sampling and right at that point would be proper sampling.

In frequency domain image, proper sampling (high SNR) would look like this:

Faintest part would just touch edges of the image.

In your original image above - it is really hard to tell where signal ends and noise begins.

From just looking at the image - it could be inner circle, but to my eye there is still some gradient between two circles and it seems that circle extends beyond image (this would indicate under sampling).

However, due to SNR - I can't be really certain. Maybe circle extends even further and that faint glow we see in corners is still signal?

By the way - circular signature in images above is from aperture and airy disk. When we enlarge image in software - signature is a bit different as transform is done in digital domain.

I think I'm detecting a hint of it in your enlarged image:

Here outline tends to be more square shaped then round:

You can check this by simply doing a noise image and enlarging it by factor of 2

Top is small image FFT and original image, and bottom is enlarged (Cubic O-Moms interpolation) and its FFT (fft is to the left).

This square will operate on both noise and original signal, so the rest of the image in high frequencies will be noise free and darker - as it is shown in FFT of your enlarged image.

 

[vlaiv]

@inFINNity Deck

Here is interesting experiment if you are up for it.

Generate random noise and print it on a piece of paper. Place that piece of paper at some distance (at least 30-40 meters to avoid too much issues from close focus) and use barlow that can be adjusted to sensor distance (you can also print a scale on same paper for reference on how much magnification each position is).

Take narrow band filter like OIII that is at 500nm and then use F/ratio that is pixel size x3, x4, x5 and x6 for example to record that noise patch. Use longer exposure to get good SNR (don't bother with stacking - if everything is ok, there should be no seeing effects in this close setup).

This way we should get very clear cut off point for aperture in frequency domain.

[neil phillips]

4 hours ago, vlaiv said:

Could it be that:

a) people associate larger image with more quality (selfish gene, wanting bigger / better/ more or comparing one self with Hubble images)

b) If person A is making such a large image, why can't I?  (competitiveness)

c) General public's expectations and trying to get likable image?

I'm not saying that any of them is doing it consciously - I just know that I found my myself on several occasions in struggle of what is expected vs "what is right" and that tells me that the struggle is real.

Of course, it could be one of those reasons. But seems a bit presumptuous. For all of them to not be testing different focal lengths Seems unlikely to me. 

And if a better result was being got at correct sampling and competitiveness was the reason for the large images. Wouldn't they get a better result just resampling the image by a factor of 2x after such said better result was captured? at correct sampling? Why would they hinder their images ? makes little sense to me ? 

[inFINNity Deck]

33 minutes ago, vlaiv said:

In your original image above - it is really hard to tell where signal ends and noise begins.

From just looking at the image - it could be inner circle, but to my eye there is still some gradient between two circles and it seems that circle extends beyond image (this would indicate under sampling).

However, due to SNR - I can't be really certain. Maybe circle extends even further and that faint glow we see in corners is still signal?

 

I took the original G-channel and did analysis on that without cropping:

I can see at least three areas: the centre, then a narrow darker region, followed by a broader and brighter one (I may have drawn it slightly too small). It appears as if in the far corners the signal get stronger again, which could be even a fourth area of signal/noise. So, if the corners still contain signal instead of noise, then the image is undersampled at f/20, otherwise it would be oversampled (which I think it is).

24 minutes ago, vlaiv said:

@inFINNity Deck

Here is interesting experiment if you are up for it.

Generate random noise and print it on a piece of paper. Place that piece of paper at some distance (at least 30-40 meters to avoid too much issues from close focus) and use barlow that can be adjusted to sensor distance (you can also print a scale on same paper for reference on how much magnification each position is).

Take narrow band filter like OIII that is at 500nm and then use F/ratio that is pixel size x3, x4, x5 and x6 for example to record that noise patch. Use longer exposure to get good SNR (don't bother with stacking - if everything is ok, there should be no seeing effects in this close setup).

This way we should get very clear cut off point for aperture in frequency domain.

That would be a great experiment, but sadly enough not feasible from my observatory.

Nicolàs

[vlaiv]

1 minute ago, inFINNity Deck said:

I took the original G-channel and did analysis on that without cropping:

That is very strange - there should be no change with cropping.

Response in frequency domain does not change based on size of the image.