53 replies to this topic

[Jan_Fremerey]

And I have not seen any video yet.

So I'm puzzled as to your line of reasoning—and even what you are trying to convey.

Our submissions overlapped. Here is part of my Mars video from yesterday night. My analyses so far are for monochrome cameras. My goal is actually to show that in video astronomy you can get high resolution images at down to f/D = 2*p/µm camera resolution.

 

CS Jan
 

Edited by Jan_Fremerey, 09 February 2023 - 01:18 PM.

[BQ Octantis]

It is quite straightforward to digitally upsample with reasonable results, even with bicubic interpolation. But what do you mean by "high resolution"?

 

"Video astronomy" generally refers to live stacking, but you're not upscaling your video. So do you mean basic alignment and rejection stacking of an AVI capture to produce a still image?

 

And what does Nyquist have to do with 2*px/µm?

[Jan_Fremerey]

But what do you mean by "high resolution"?

"Video astronomy" generally refers to live stacking

And what does Nyquist have to do with 2*px/µm?

High resolution imaging is close to the resolving power of the telescope optics. This, in fact, due to poor seeing conditions does not apply in case of my above camera and drizzle grid demonstration. Did you have a look at the original video?

 

In the present discussion I do not restrict "video astronomy" to live stacking.

 

The established recommendation of f/D > 3.5*p/µm for camera adaptation is generally justified by experts with reference to Nyquist. On the basis of practical experience I find f/D > 2*p/µm and propose destrucive drizzle stacking of the camera grid pattern as actual reason for the lower limit.
 

CS Jan

[BQ Octantis]

I did look at the video. And I zoomed way in during playback with whatever live video scaling interpolation VLC uses.

 

Are you saying that you can sample at 2*p/µm with your mono camera on your 10-in Newt because you can upsample with drizzle and/or sinc to the practical detail limit of the scope with results you find esthetically acceptable?

[Jan_Fremerey]

Are you saying that you can sample at 2*p/µm with your mono camera on your 10-in Newt ... with results you find esthetically acceptable?

Yes, I do - with special reference to examples displayed on the upper part of my website.

 

CS Jan

Edited by Jan_Fremerey, 10 February 2023 - 04:06 AM.

[Jan_Fremerey]

Are you saying that you can sample at 2*p/µm with your mono camera on your 10-in Newt because you can upsample with drizzle and/or sinc to the practical detail limit of the scope with results you find esthetically acceptable?

Assuming the Theophilus/Cyrillus region on the upper part of my website is still "esthetically acceptable", I have just done a straightforward analysis of the resolving power achieved with the photograph that was taken by video at f/D = 2.05*p/µm.

 

 

 

The diagram reflects the grayscale profile along the tiny line across part of the craterlet chain on the northern Cyrillus wall. Image resolution is 0.1"/pixel, so shadow centers of adjacent craterlets are 0.7" apart. Obviously, this is still quite well above the 0.5" Rayleigh limit of my 10-in mirror.

 

CS Jan

Edited by Jan_Fremerey, 14 February 2023 - 03:28 AM.

[Jan_Fremerey]

demonstrate an outstanding benefit of video astronomy: the effective extinction of camera grids by destructive drizzle stacking, ...

You may easily demonstrate the "dynamic" extinction of a foreground grid structure by putting a sieve in front of a newspaper:

 

 

When moving the screen around, you can clearly read the writing underneath.

 

CS Jan 

Edited by Jan_Fremerey, 19 February 2023 - 08:05 AM.

[BQ Octantis]

In 27 years of doing digital photography, I have never seen anyone claim that their sensor grid lines (i.e., the spaces between their pixels) limited their detail resolving ability.

Edited by BQ Octantis, 19 February 2023 - 12:14 PM.

[Jan_Fremerey]

In 27 years of doing digital photography, I have never seen anyone claim that their sensor grid lines (i.e., the spaces between their pixels) limited their detail resolving ability.

According to Section 6.1 of "Planetary Imaging FAQ" latest edidion, pixel pitch is crucial.

 

CS Jan

[BQ Octantis]

And now we're going in circles.

[Jan_Fremerey]

I have never seen anyone claim that their sensor grid lines (i.e., the spaces between their pixels) limited their detail resolving ability.

In the above photograph with the non-moving grid, can YOU read the writing behind ? When taking astro images by video, we always have relative movement between camera grid and image detail. Of course, nobody would claim limited detail by "spaces between their pixels", simply because these structures will vanish by the stacking procedure.

 

We should be aware that in video astronomy the stacking procedure enhances image detail by realignment, and thereby supresses both, seeing induced blur of image detail as well as border lines between adjacent camera pixels.

Edited by Jan_Fremerey, 20 February 2023 - 10:21 AM.

[BQ Octantis]

In the above photograph with the non-moving grid, can YOU read the writing behind ? When taking astro images by video, we always have relative movement between camera grid and image detail. Of course, nobody would claim limited detail by "spaces between their pixels", simply because these structures will vanish by the stacking procedure. 

I do the same looking into my microwave oven by just moving my head around. But it's an unrelated model. The screen obscures the text, but the parts that show are resolved to much greater detail than the screen gap size.

 

In the case of a Bayer pattern, there are indeed many spaces that normally get interpolated with demosaicing. Drizzle does what you are proposing to fill in the gaps. But in the case of your mono camera, there is no space between the pixels. A single sub has no gaps—you don't need to stack.

[Jan_Fremerey]

But it's an unrelated model.

Unfortunately, our posts overlapped so you could not read my two additional lines.

[Jan_Fremerey]

And now we're going in circles.

Just to make clear what I'm trying to say I refer to my recent Mars image that was taken at f/D = 2.05*p/µm,

 

 

and here is an animated view of 6 consecutive frames from the original video that have been taken at 1.8 ms and 555 fps:

 

 

You see the camera grid and the planet image jiggling around thereon. It is obvious that the camera grid is much fainter than the planetary detail. As the stacking software realines the planet before summing up, you may imagine that the tiny grid striucture is shifted around by the alignment process and will be wiped out by incoherent stacking.

 

CS Jan

Edited by Jan_Fremerey, 21 February 2023 - 05:49 AM.

[BQ Octantis]

So you're saying stacking makes the image clearer?

[Jan_Fremerey]

Yes, of course, due to alignment and stacking, seeing-related blur and camera raster are largely eliminated. That is the essential benefit of video astronomy.  Do you see it differently ?

[BQ Octantis]

We stack to leverage the law of large numbers, so that the value of each pixel in the averaged stack is closer to the mean luminosity off the target than the value of the pixel in an individual sub.

 

But seeing-related blur is actually worsened by stacking. So we also stack for SNR so we can undo the wavefront spreading via deconvolution and/or wavelets—in the hopes that our processing can recover more detail than the noise it also enhances.

 

The camera "raster" (pixel pitch) is unchanged unless you resample. As I already mentioned, there are many methods to upsample; Drizzle is a method of upsampling by first subsampling (i.e., forcing gaps between pixels) and then taking advantage of the natural dithering you show in your animation to fill in the gaps.

 

Your sample animation that shows the pixels as large blocks used nearest-neighbor interpolation for upsampling. Had you not upsampled, the pixel pitch and the screen pitch would be 1:1.

 

That is how I see it.

Edited by BQ Octantis, 23 February 2023 - 08:16 PM.

[Jan_Fremerey]

But seeing-related blur is actually worsened by stacking.

Please note that stacking is done with image subsection content after alignment to its most frequent position. So image detail is enhanced by the stacking procedure rather than blurred.
 

[BQ Octantis]

Have you done a stacking experiment to confirm your belief?

 

Compare the following aligned stacks of your data (which are increments of 1 stop of SNR) without enhancement:

 

1 sub

4 subs

16 subs

64 subs

256 subs

1024 subs

4096 subs

 

Then report back with results in a table similar to this:

 

 

BQ

Edited by BQ Octantis, 24 February 2023 - 05:56 PM.

[RedLionNJ]

I've been trying to follow along every day or so, but am not seeing any convincing arguments for your proposition, Jan.

 

Might it be possible to capture two consecutive videos (thus a best attempt at comparable seeing), one at prime focus and one with a barlow, and compare the stacked/sharpened content?

[Jan_Fremerey]

Then report back with results in a table similar to this:

Thanks a lot for your impressive demonstration ! You exactly hit my point: reducing focal lenght enables shorter shutter times and increased fps for better suppression of seeing-related image blur and signal noise. Going back from 4.45 to 2 ppAd enables 5x increase in fps.

 

CS Jan
 

[BQ Octantis]

Thanks a lot for your impressive demonstration ! You exactly hit my point: reducing focal lenght enables shorter shutter times and increased fps for better suppression of seeing-related image blur and signal noise. Going back from 4.45 to 2 ppAd enables 5x increase in fps.

Not on my Canon 600D/T3i. Its frame rate is fixed at ~9-10 fps, its shutter is fixed at 1/30th second. I shoot Jupiter, Mars, and Venus at ~9 ppAd for several reasons (for Jupiter and Mars for the higher image quality and more forgiving depth-of-field from the 12.5mm eyepiece, for Venus because I can't shoot any faster without clipping because of the camera's fixed shutter and minimum ISO); Saturn I vacillate between 4 and 9, depending on seeing (trading SNR and image quality). But I present at ~5 ppAd because that is the esthetic limit I've found for a fixed image from my Mak 180.

 

But we aren't talking about my DSLR or my Mak 180. We're talking about your mono captures from your 10-in Newt. And without some sort of ground truth reference for comparison, your trade comparison seems to be between upsampling artifacts and sharpening artifacts.

 

BQ

[Jan_Fremerey]

Might it be possible to capture two consecutive videos

In my initial post I try it with one and the same video by retroactively reducing camera resolution fom 2.93 to 2.05 and 1.37 pixels per Airydisk.
 

[Jan_Fremerey]

And without some sort of ground truth reference for comparison, your trade comparison seems to be between upsampling artifacts and sharpening artifacts.

I kindly invite you to analyse my example images for artifacts and find "some sort of ground truth reference for comparison" at your discretion.

 

CS Jan

[BQ Octantis]

I kindly invite you to analyse my example images for artifacts and find "some sort of ground truth reference for comparison" at your discretion.

 

I did.

 

I do not have a database of Hubble images from the same moment for comparison.

 

Without that, I can't discern processing artifact from planetary feature. So there is no basis to judge one method from another.

 

Moreover, as I already stated, sampling simulations are just that: sampling simulations. They are no substitute for actually comparing two methods.

Edited by BQ Octantis, 24 February 2023 - 08:03 PM.