Help my brain cells : Planetary camera considerations

[inFINNity Deck]

Hi Geoff,

here is an example of how sampling would work at different f-ratios. The image that I used was the luminance recording of Jupiter on 9 November 2022 at around 20:51UTC. Here is the final processed image:

The recording was done at f/20 using a ZWO ASI174MM, TeleVue 2x PowerMate and C11 EdgeHD. The camera would require 5.9 x 3.7 = f/21.8.

In the following image I took the luminance channel and ran it through ImageJ (left set). Then I resized it by 50% and again resized it 200% (middle set), which would have been roughly the same had I imaged the planet at f/10 and then resized it 200%. Indeed it looks a bit softer than the original, but after a little bit of sharpening (right hand set) we get an image that is visually not far off from the original.

The frequency spectrum shows that resizing the image caused a little bit loss of data in the higher frequencies (so smaller details, the corners are black now), but most of the information was preserved. Sharpening added some detail, but that would have been no real detail, but mainly magnified noise.

That hardly any detail was lost is mainly due to seeing as that severely affects image quality. Preferably we would like to image under diffraction limited circumstances, but at this aperture (11" or 279.4mm) that would require a seeing of below 0.4 arc-seconds, which is very rare under our moderate climate here in the Netherlands and UK:

If you want to know more about how seeing affects detail, you may want to read the following article (this time written in English, you may wish to skip the first six pages): https://arxiv.org/abs/2208.07244

Nicolàs

[geoflewis]

1 hour ago, vlaiv said:

I'm not sure if I actually saw example of evidence to support this.

Could you post an example that shows your findings, or at least explain what you find better in over sampled images (like additional detail revealed or anything)?

Here are 2 similar views of Mars taken several weeks apart the one dated 13 Nov 2022, was at F13 (no Barlow), the one dated 20 Dec was with the barlow in poor seeing captured through passing cloud, but I still prefer it. A much better, but slightly different view was captured with the barlow on 16 Dec. One noticible thing is that the edge rind is much more prominent in all my 'correctly' sample images. You can see them all on my Astrobin page here

I pretty much changed from F13 (without the barlow) to F21 (with the barlow) wef 2 Dec. Here are 2 similar views of Mars one from 1 Dec at F13 and the second on 2 Dec at F21 with the barlow. I know which I prefer. I have also included the TIFFs if you want to poke around in imageJ or whatever you use.

2022-12-01-2345_8-GDL-RGB-Mars_lapl4_ap5_P30_sharp90_R6(Hist50-200)_Wav(10-20-30).tif

2022-12-02-0007_5-GDL-RGB-Mars_lapl4_ap26_P30_sharp90_R6(Hist50-200)_Wav(1-1-20-30-40-50).tif

I have not used drizzle or resampled, resized either of those last 2 images, just run through AS3! and applied wavelets in Rexistax. If you want the raw stack TIFFs out of AS3!, here they are....

2022-12-01-2345_8-GDL-RGB-Mars_lapl4_ap5_P30_sharp90.tif

2022-12-02-0007_5-GDL-RGB-Mars_lapl4_ap11_P30_sharp90.tif

I will be genuinly interested in your findings and explanations please.

[geoflewis]

3 minutes ago, inFINNity Deck said:

Hi Geoff,

here is an example of how sampling would work at different f-ratios. The image that I used was the luminance recording of Jupiter on 9 November 2022 at around 20:51UTC. Here is the final processed image:

The recording was done at f/20 using a ZWO ASI174MM, TeleVue 2x PowerMate and C11 EdgeHD. The camera would require 5.9 x 3.7 = f/21.8.

In the following image I took the luminance channel and ran it through ImageJ (left set). Then I resized it by 50% and again resized it 200% (middle set), which would have been roughly the same had I imaged the planet at f/10 and then resized it 200%. Indeed it looks a bit softer than the original, but after a little bit of sharpening (right hand set) we get an image that is visually not far off from the original.

The frequency spectrum shows that resizing the image caused a little bit loss of data in the higher frequencies (so smaller details, the corners are black now), but most of the information was preserved. Sharpening added some detail, but that would have been no real detail, but mainly magnified noise.

That hardly any detail was lost is mainly due to seeing as that severely affects image quality. Preferably we would like to image under diffraction limited circumstances, but at this aperture (11" or 279.4mm) that would require a seeing of below 0.4 arc-seconds, which is very rare under our moderate climate here in the Netherlands and UK:

If you want to know more about how seeing affects detail, you may want to read the following article (this time written in English, you may wish to skip the first six pages): https://arxiv.org/abs/2208.07244

Nicolàs

Thanks Nicolàs. Unfortunately, I do not understand a lot of what you did, nor similar analysis done by @vlaiv. All I do is try my gear and see what works for me and IMHO what I get from oversampled data is better than what I get from 'correctly' sampled data. This may just be an eyesight thing, with me prefering to work with a larger image, but I can't change my eyes, whereas I can change the image size with a barlow in the optical train.

Got to run to a medical appointment now, so I'll look out for any further answers when i get back in an hour or so.

[vlaiv]

13 minutes ago, geoflewis said:

I will be genuinly interested in your findings and explanations please.

You say that you prefer left one out of these two:

why?

(right one is sampled at F/13 and just upsampled to match the size of left one).

I can see some features in the left image that are more distinct than in right image, but I can also see features in the right image that are more distinct than in left - which leads me to conclude that it is just seeing and stacking variation and difference in processing between the two.

Is there any particular feature that makes left image better that we can contribute to increase in sampling?

[inFINNity Deck]

Here is the analysis of the 20 December image. I have split the image into RGB, then processed each channel using ImageJ:

I have not resized the images, only show them here at 50% to make them fits the screen.

Clearly the FFT shows that the images are well oversampled. No detail is lost when reducing it to 25% and resizing it 400% again:

Nicolàs

[Altocumulus]

I just did a search in the plugin library. It was the first on offer.

 

 

 

 

 

 

Edited December 23, 2022 by Altocumulus

[johnturley]

2 hours ago, geoflewis said:

Thanks Vlaiv, I know that you’ve said it many times, but my personal experience (and I think many others) says that oversampling gets better results. I have tried both correct sampling and over sampling and my oversampled images are almost always the better ones. I don’t understand why, maybe it easier to obtain accurate focus with larger over sampled image on screen, or something else, but operating at F21 or F24 gives me better results despite ‘correct’ sampling is ~F14. I cannot deny the maths/physics that you describe, but equally I cannot deny the evidence of my own eyes.

Depending on your system, you may also need to oversample and at a higher focal ratio than the theory suggests, to get a decent image size.

John 

[inFINNity Deck]

2 hours ago, geoflewis said:

Thanks Nicolàs. Unfortunately, I do not understand a lot of what you did, nor similar analysis done by @vlaiv. All I do is try my gear and see what works for me and IMHO what I get from oversampled data is better than what I get from 'correctly' sampled data. This may just be an eyesight thing, with me prefering to work with a larger image, but I can't change my eyes, whereas I can change the image size with a barlow in the optical train.

Basically what we are showing is that your image is oversampled. That means that you could use a lower f-ratio and then blow up the image to get the size you want. That in turn has the advantage that the exposure times can be much shorter, which in turn may result in more detail as the capturing will be less affected by seeing.

Nicolàs

Edited December 23, 2022 by inFINNity Deck
typo

[vlaiv]

4 minutes ago, johnturley said:

Depending on your system, you may also need to oversample and at a higher focal ratio than the theory suggests, to get a decent image size.

John 

You can get equally decent image size by taking properly sampled image and just upscaling it - look at my post above.

I personally prefer smaller but sharper image than large blurry one.

[vlaiv]

Just for reference - here is what level of data can be had at what sampling rate and comparison (I tried to make very high resolution reference image resemble capture as much as possible without loss of detail):

Level of detail in properly sampled image looks much closer to the level of detail that can maximally be held in that sampling rate.

Difference in level of detail in larger image is greater.

 

[johnturley]

11 minutes ago, vlaiv said:

You can get equally decent image size by taking properly sampled image and just upscaling it - look at my post above.

I personally prefer smaller but sharper image than large blurry one.

Which program do you recommend for upscaling, I tried upscaling some images in GIMP, but the results were not that great. 

Also if the initial image is too small you may struggle to get a sufficient array of alignment points in programs such as AutoStakkert and Registax.

John 

[vlaiv]

17 minutes ago, johnturley said:

Which program do you recommend for upscaling, I tried upscaling some images in GIMP, but the results were not that great. 

If you want to resample raw data (linear) - then ImageJ with TransformJ gives large number of resampling algorithms.

If you want to resample processed image - use IrfanView - it has Lanczos resampling (that is what I used to compare the two Mars images above at the same scale).

18 minutes ago, johnturley said:

Also if the initial image is too small you may struggle to get a sufficient array of alignment points in programs such as AutoStakkert and Registax.

I think that alignment point size should be matched to seeing conditions rather than to size of the target.

If seeing quivering is small and rapid - one should use small alignment points but if it is large in magnitude or if seeing is without larger variations of tilt (just softening of the image without quivering) - then it is ok to use larger alignment points.

[Space Cowboy]

I think often people over sharpen lower sampled images which produces more artifacts and thickening of detail.

Personally I find a larger image easier to focus during capture but obviously seeing conditions dictate how high a f ratio I would go. 

 

[geoflewis]

1 hour ago, vlaiv said:

You say that you prefer left one out of these two:

why?

(right one is sampled at F/13 and just upsampled to match the size of left one).

I can see some features in the left image that are more distinct than in right image, but I can also see features in the right image that are more distinct than in left - which leads me to conclude that it is just seeing and stacking variation and difference in processing between the two.

Is there any particular feature that makes left image better that we can contribute to increase in sampling?

Thanks Vlaiv, I agree that they are similar, so it may just be what I said several times, which is that I prefer to work with the bigger image on screen during capture. I know that I can zoom the FireCapture window, but I find too much zoom in FC gives me a poorer quality image, so accurate focus becomes more challenging. I know that I can resize the images in post processing, but it is getting a good focus at time of capture which is critical and the bigger over sampled image helps me.

[geoflewis]

1 hour ago, inFINNity Deck said:

I have not resized the images, only show them here at 50% to make them fits the screen.

Clearly the FFT shows that the images are well oversampled. No detail is lost when reducing it to 25% and resizing it 400% again:

Thanks Nicolàs, I do not know what FFT is, or what it is that is clearly showing that the images are well oversampled. It is not at all clear to me what shows this, so where can I see that please? Also I think that the RESIZED 25% - 400% image has lost a bit of detail, though not a lot - maybe it just looks smoother to me.

As I said to Vlaiv, the biggest issue for me is the size of the image on screen when I am trying to focus, prior to capture.

[geoflewis]

1 hour ago, vlaiv said:

Just for reference - here is what level of data can be had at what sampling rate and comparison (I tried to make very high resolution reference image resemble capture as much as possible without loss of detail):

Level of detail in properly sampled image looks much closer to the level of detail that can maximally be held in that sampling rate.

Difference in level of detail in larger image is greater.

 

Sorry Vlaiv, I don't understand what you did here, so please could you explain some more, thank you.

[vlaiv]

Just now, geoflewis said:

Sorry Vlaiv, I don't understand what you did here, so please could you explain some more, thank you.

I didn't do anything particular except that I took high resolution reference image that shows same features as your two images.

Originally image is very high resolution as can be seen from this example:

When you take such image and scale it to particular sampling rate - you get what is maximum amount of detail that can be recorded at that sampling rate (since there is no blur from optics to speak of).

This is useful when you want to compare your own image at certain resolution with what can be recorded at that resolution. If your image looks more blurry than what can be recorded at particular resolution and can't be sharpened further - that simply means that it is over sampled. It is lacking the detail that corresponds to that image size.

I made such comparison with two of your images - one at F/13 and one at F/21.  I did a little of processing on reference image - just desaturation and brightness to try to match it visually to your images. I kept the level of detail intact.

What can be seen from this comparison is how much your images are lacking detail compared to max detail that can be recorded at certain image size.

To my eye - smaller image is more similar to reference image at that scale than larger image - or in another words larger image lacks more detail compared to what can be recorded at larger size. To again put it differently - smaller image is less over sampled that larger image.

In any case, I believe that it is good thing to be aware what properly sampled image at certain image size looks like.

[vlaiv]

Just to clarify further what I mean by above - look at this:

At this scale images look the same - high resolution image simply does not look high resolution any more compared to your image as sampling has removed any additional detail compared to your image. Now they are matched by level of detail.

When this happens - you have reached the point of optimum sampling for your data (if we judge by visual part alone).

[geoflewis]

13 minutes ago, vlaiv said:

I didn't do anything particular except that I took high resolution reference image that shows same features as your two images.

Originally image is very high resolution as can be seen from this example:

When you take such image and scale it to particular sampling rate - you get what is maximum amount of detail that can be recorded at that sampling rate (since there is no blur from optics to speak of).

This is useful when you want to compare your own image at certain resolution with what can be recorded at that resolution. If your image looks more blurry than what can be recorded at particular resolution and can't be sharpened further - that simply means that it is over sampled. It is lacking the detail that corresponds to that image size.

I made such comparison with two of your images - one at F/13 and one at F/21.  I did a little of processing on reference image - just desaturation and brightness to try to match it visually to your images. I kept the level of detail intact.

What can be seen from this comparison is how much your images are lacking detail compared to max detail that can be recorded at certain image size.

To my eye - smaller image is more similar to reference image at that scale than larger image - or in another words larger image lacks more detail compared to what can be recorded at larger size. To again put it differently - smaller image is less over sampled that larger image.

In any case, I believe that it is good thing to be aware what properly sampled image at certain image size looks like.

10 minutes ago, vlaiv said:

Just to clarify further what I mean by above - look at this:

At this scale images look the same - high resolution image simply does not look high resolution any more compared to your image as sampling has removed any additional detail compared to your image. Now they are matched by level of detail.

When this happens - you have reached the point of optimum sampling for your data (if we judge by visual part alone).

I think I now understand......, a least a little better than I did....🤔🙄

 

[Altocumulus]

2 hours ago, inFINNity Deck said:

Here is the analysis of the 20 December image. I have split the image into RGB, then processed each channel using ImageJ:

I have not resized the images, only show them here at 50% to make them fits the screen.

Clearly the FFT shows that the images are well oversampled. No detail is lost when reducing it to 25% and resizing it 400% again:

Nicolàs

What happens analysis-wise if one of those images (or a live image) is de-focused?

[vlaiv]

12 minutes ago, Altocumulus said:

What happens analysis-wise if one of those images (or a live image) is de-focused?

Defocus is another level of blur that adds up - it reduces effective resolution of the image and makes it harder to sharpen it up (it attenuates high frequencies even more).

 

[Altocumulus]

7 minutes ago, vlaiv said:

Defocus is another level of blur that adds up - it reduces effective resolution of the image and makes it harder to sharpen it up (it attenuates high frequencies even more).

 

Taking that to a previous question, then - if the sampling ratio f and pixel size isn't correct, could that effect the ability to achieve focus easily. I know conditions weren't good the last twice I tried Jupiter/Mars but I just couldn't get focus - not even a hint in lucky imaging - and the 174 was motoring at ~100 fps.

[inFINNity Deck]

 

1 hour ago, geoflewis said:

Thanks Nicolàs, I do not know what FFT is, or what it is that is clearly showing that the images are well oversampled. It is not at all clear to me what shows this, so where can I see that please?

FFT stands for Fast Fourier Transformation, which takes the data from the spatial domain (the image) to the frequency domain (the Frequency Spectra (scatter-plots) below the images that show which frequencies are in the image). The frequency spectra are scaled to the same size as the original images. If the original image is 100% diffraction limited sampled, the corresponding frequency spectrum would show data all over it. Below image from my article was generated using a white-noise image that was sampled optimally (far left). The frequency spectrum below it shows a very similar noisy image that is completely filled with data, which indicates that indeed it is at least not undersampled. The image in the centre here below is that same white-noise image but with a black border around it. The frequency spectrum of that again is completely filled as the original image is properly sampled. At the far right I have taken the original image and resized it 200%. The frequency spectrum of that shows a black border, indicating that the 200% resized image is oversampled.

The smaller image at the lower left, the one with the red circles, shows that the input image is not 100% white noise as within the red circles the response is not random. These two areas are also visible in the other two spectra.

 

1 hour ago, geoflewis said:

Also I think that the RESIZED 25% - 400% image has lost a bit of detail, though not a lot - maybe it just looks smoother to me.

It should look a bit smoother as some averaging is done of course by reducing it to 25%.

 

1 hour ago, geoflewis said:

As I said to Vlaiv, the biggest issue for me is the size of the image on screen when I am trying to focus, prior to capture.

When I started planetary imaging I tried visual focusing, but never got good results apart from by chance. Now I use a Bahtinov mask on a nearby star or, in case of Jupiter, on one of the moons. During focusing I raise exposure time to 100-500ms to get a more average result of the diffraction pattern, and I zoom the live image to about 200%. Using it at a low exposure time will make the diffraction pattern to change too much and very difficult to properly analyse.

Nicolàs

 

Edited December 23, 2022 by inFINNity Deck

[geoflewis]

16 minutes ago, inFINNity Deck said:

 

FFT stands for Fast Fourier Transformation, which takes the data from the spatial domain (the image) to the frequency domain (the Frequency Spectra (scatter-plots) below the images that show which frequencies are in the image). The frequency spectra are scaled to the same size as the original images. If the original image is 100% diffraction limited sampled, the corresponding frequency spectrum would show data all over it. Below image from my article was generated using a white-noise image that was sampled optimally (far left). The frequency spectrum below it shows a very similar noisy image that is completely filled with data, which indicates that indeed it is at least not undersampled. The image in the centre here below is that same white-noise image but with a black border around it. The frequency spectrum of that again is completely filled as the original image is properly sampled. At the far right I have taken the original image and resized it 200%. The frequency spectrum of that shows a black border, indicating that the 200% resized image is oversampled.

The smaller image at the lower left, the one with the red circles, shows that the input image is not 100% white noise as within the red circles the response is not random. These two areas are also visible in the other two spectra.

 

It should look a bit smoother as some averaging is done of course by reducing it to 25%.

 

When I started planetary imaging I tried visual focusing, but never got good results apart from by chance. Now I use a Bahtinov mask on a nearby star or, in case of Jupiter, on one of the moons. During focusing I raise exposure time to 100-500ms to get a more average result of the diffraction pattern, and I zoom the live image to about 200%. Using it at a low exposure time will make the diffraction pattern to change too much and very difficult to properly analyse.

Nicolàs

 

Thanks for the further explanation Nicolàs, so are you saying it is the large black border in the FFT  of my Mars images that tells you it is oversampled?

Regarding using a B-Mask for planetary focusing, for me that is a complete no. Even with aftermarket mirror locks on my C14, moving the scope from a bright star back to the planet will likely see the mirror move enough to lose focus, but in any case I don't have a B-mask for the C14..... I did have one for my old 10" Meade LX200, but that never worked well enough for me for planetary imaging, so I almost always use planetary surface features to focus.

[vlaiv]

58 minutes ago, Altocumulus said:

Taking that to a previous question, then - if the sampling ratio f and pixel size isn't correct, could that effect the ability to achieve focus easily. I know conditions weren't good the last twice I tried Jupiter/Mars but I just couldn't get focus - not even a hint in lucky imaging - and the 174 was motoring at ~100 fps.

Correct focus position does not depend on either of those two things and I don't really see why would either over or under sampling prevent anyone from achieving good focus.

It's a bit like saying that finding correct focus for visual needs very high magnification. I've never had issues where I could not focus at particular magnification. I did however have situations where seeing was so poor that I was not certain if I had correct focus.