Practice with processing severely undersampled shots, M31-M33-M101

[ONIKKINEN]

Perhaps i am a little bored because it has been cloudy since October, but decided to try processing some purposefully undersampled and underintegrated shots to see if they can be worked with to the same extent as a full integration with less binning would.

First, 4 panel mosaic of M31 with just 1 60s sub per panel at effectively BINx12:

Honestly surprised how much comes out of just 25 photons of median signal per pixel at capture resolution.

Then M33 which i found was not workable as a single sub, probably because 60s is not enough to get any number of photons from some parts of the galaxy so this time it is 5x 60s subs again at BINx12:

Again very surprised, looks about as good as a full integration would, just smaller and less detailed.

Then an M101 i captured in awful full Moon and 3''+ FWHM seeing conditions last winter. Back then it produced a terrible image at (i think) 3h integration. This time binned it x6 and i actually like it a lot:

Mostly did this to have a real life test on the benefits of binning as its often just quoted (by myself too) that binning x3 will improve SNR by a factor of 3 etc. Looks to be the case from the results which is hardly shocking but still somewhat surprising to see that just 1 or 5 minutes produces a "finished" image with undersampling taken to an extreme.

[vlaiv]

It also shows that large scopes are excellent wide field instruments

Level of detail at that scale is beyond anything lens or small focal length scope could capture.

[ONIKKINEN]

3 minutes ago, vlaiv said:

It also shows that large scopes are excellent wide field instruments

Level of detail at that scale is beyond anything lens or small focal length scope could capture.

The experiment with M31 did give me a crazy idea. If just 1 minute at Bin12 gets what i think is already a decent image, maybe the same logic could extend to a really large target that i have not even considered before. Like the entire veil nebula for example that would require something like a 4x4 mosaic with 1m focal length and APS-C sensor size. Total integration would still be short so could be done in one good night. Might have to try that (if i ever get a clear night again).

[vlaiv]

15 minutes ago, ONIKKINEN said:

The experiment with M31 did give me a crazy idea. If just 1 minute at Bin12 gets what i think is already a decent image, maybe the same logic could extend to a really large target that i have not even considered before. Like the entire veil nebula for example that would require something like a 4x4 mosaic with 1m focal length and APS-C sensor size. Total integration would still be short so could be done in one good night. Might have to try that (if i ever get a clear night again).

Thing is - large telescope with mosaic and binning is "as fast" as small telescope for same target resolution.

Here is simple way to understand it.

Imagine you have 4" telescope and 8" telescope. Let them both be F/5 for the sake of argument (easy calculations. This means that focal length of 8" is double that of 4" (same as aperture).

Let's further say that 4" telescope spends 4 hours on target.

You'll need 4 panels to cover same FOV with 8" telescope, so you can spend only 1 hour for each panel to get same total imaging time.

8" gathers x4 more light, but since focal length is twice as long - "pixel surface" (sky covered by pixel) is 1/4 that of 4" (sampling rate is halved). Two cancel and 8" is as fast as 4" - but then you bin x2 - and 8" now becomes x4 faster.

So each 1h panel will have same SNR as 4h part of image with 4".

You end up with same speed between two systems - except the fact that 8" will produce sharper image because of Airy disk component of final FWHM.

It is indeed worth doing wide field mosaics with larger scope - it will be equally fast as with smaller instrument (if you target same sampling rate) - or even faster if you go for lower sampling rate, which is fine for wide field - as you've seen - nothing bad comes from under sampling (no square stars and such ).

 

[ONIKKINEN]

5 minutes ago, vlaiv said:

Thing is - large telescope with mosaic and binning is "as fast" as small telescope for same target resolution.

Here is simple way to understand it.

Imagine you have 4" telescope and 8" telescope. Let them both be F/5 for the sake of argument (easy calculations. This means that focal length of 8" is double that of 4" (same as aperture).

Let's further say that 4" telescope spends 4 hours on target.

You'll need 4 panels to cover same FOV with 8" telescope, so you can spend only 1 hour for each panel to get same total imaging time.

8" gathers x4 more light, but since focal length is twice as long - "pixel surface" (sky covered by pixel) is 1/4 that of 4" (sampling rate is halved). Two cancel and 8" is as fast as 4" - but then you bin x2 - and 8" now becomes x4 faster.

So each 1h panel will have same SNR as 4h part of image with 4".

You end up with same speed between two systems - except the fact that 8" will produce sharper image because of Airy disk component of final FWHM.

It is indeed worth doing wide field mosaics with larger scope - it will be equally fast as with smaller instrument (if you target same sampling rate) - or even faster if you go for lower sampling rate, which is fine for wide field - as you've seen - nothing bad comes from under sampling (no square stars and such ).

 

That does make sense regarding total time required and such. Also, figured i might try M42. It never rises higher than 25 degrees here so i have just not bothered as it will be soft due to the extreme dispersion this low in the sky. But why would i care about some star size issues if the presentation resolution is at 9'' per pixel? No way the sky messes things that far so it should be a decent shot with some effort from a location that has little light pollution (the other reason why 25 degrees is a terrible idea).

On the square stars thing, it gets thrown around sometimes in discussion but not sure i have ever actually seen a square star so that must just be one of those things that gets quoted because someone else quoted it. I guess there are differences in how some image viewing software do interpolation with zooming but what i would assume to be the bare minimum - Windows image viewer - shows no square stars at 500% zoom on the Bin x6 M101:

Doesn't look good, but doesn't look square either. Even the diffraction spikes have survived the 500% zoom somehow and do not look particularly aliased.

[vlaiv]

18 minutes ago, ONIKKINEN said:

On the square stars thing, it gets thrown around sometimes in discussion but not sure i have ever actually seen a square star so that must just be one of those things that gets quoted because someone else quoted it. I guess there are differences in how some image viewing software do interpolation with zooming but what i would assume to be the bare minimum - Windows image viewer - shows no square stars at 500% zoom on the Bin x6 M101:

Square stars are consequence of nearest neighbor interpolation when zooming in.

Look at the same image zoomed in to 500% - but with nearest neighbor interpolation:

That is the origin of the myth - most software a decade or two ago used it as default when zooming in (as it is very fast and easy to implement) - and people got impression that data looks like that, but it is actually artifact of interpolation used.

But nowadays - nearest neighbor interpolation is no longer default one - so we can rarely see the effect when zooming in.

[pipnina]

33 minutes ago, vlaiv said:

Square stars are consequence of nearest neighbor interpolation when zooming in.

Look at the same image zoomed in to 500% - but with nearest neighbor interpolation:

 

But nowadays - nearest neighbor interpolation is no longer default one - so we can rarely see the effect when zooming in.

I think it depends on the program used to view the image.

Editing programs seem to prefer NN interpolation, like RawTherapee (image 1) and PixInsight (image 2). viewing programs like windows media player and chromium browser prefer bicubic or similar smooth interpolation algorithms.

 

[tomato]

@ONIKKINEN, thanks for posting these, very impressive and interesting results, much better than I thought they they would be.

[Ratlet]

Can you give a little bit more detail on this technique?  I get that you are doing a mosaic but then my brain falls over.  Are you actually binning these?  How big is the final image?

I'd quite like to blitz a few targets on Monday as it's the last time I'll be imaging for a few weeks.  I might setup with the 130pds and do this (maybe 5 minutes per panel) as the quality looks good enough for me.

[ONIKKINEN]

4 hours ago, Ratlet said:

Can you give a little bit more detail on this technique?  I get that you are doing a mosaic but then my brain falls over.  Are you actually binning these?  How big is the final image?

I'd quite like to blitz a few targets on Monday as it's the last time I'll be imaging for a few weeks.  I might setup with the 130pds and do this (maybe 5 minutes per panel) as the quality looks good enough for me.

The only mosaic of the 3 here is the M31 as it doesn't fit on the field of view of my 1025mm fl scope in any orientation. So that one technically has 4 minutes of total integration, but any single point in the image has just 1 sub of data, so 60s to work with. The image is workable because of the SNR increase from binning (factor of 12, same thing as shooting 144x longer!), not necessarily because of the mosaicing.

But i mentioned mosiacing in another comment because that's a way to make the image size larger while still binning to a silly degree, take the image of M31 compared to M33 as an example. The M31 image is larger than M33 even if both are binned to the same level (x12) because it is mosaiced from 4 panels, but both have just 1 minute of data. The image dimensions of a Bin12 image are something like 520x320px so its a very small image, but stitch those together for a target that requires maybe 3x3 panels (M42 for my scope) or 4x4 panels (Veil nebula) and you get a reasonably sized 1500px or 2000px image.

And to clarify, nothing is binned on camera or during capture as that is wasteful. Binning on camera loses the raw data and you cant choose another level afterwards so no point in doing that. Registration will also likely just fail in extremely binned shots where the app doing star alignment doesn't find any suitable stars because they are all basically a pixel. Photometric colour calibration in Siril also doesnt seem to like very low resolutions, so for some of these i did it before binning (or binned a bit, did PCC and binned more). The workflow for these shots was: calibration first, then bayer split the OSC data into its raw mono components, which effectively also bins the image x2. Then stack the ones that were stacked or mosaiced before binning further. After that recomposited the mono stuff back to a colour image and binned the linear stacks to the final level with ASTAP (ASTAP binning works for colour data). Since the images were already binned x2 pre stacking, now i did a bin x3 to the M101 to reach its final Bin x6 form and a bin x3 and a further bin x2 (2x3x2 in total) to the M31 and M33 ones.

[geeklee]

Thanks for posting these examples @ONIKKINEN  As enjoyable to view as many "big versions" I've seen.  You still get a sense of the detail.