Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

64 Mpxl Lunar Mosaic

  • Please log in to reply
34 replies to this topic

#1 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 10:34 AM

Because I am a glutton for punishment, I tried again last night for a ~64 Mpxl lunar mosaic, and this time I succeeded. 8 Mpxl SVBONY SV205 in 10" F/4.8 Sky-Watcher Dobsonian at prime focus. Each panel is 50% of 2000 frames stacked. 15 panels, 667 GB of ser data. Batch stacked in Autostakkert, batch deconvolved in Astra Image, auto stitched in Microsoft ICE, gamma corrected in Astra Image.

 

The final result is let down by the poor seeing, but I take what I can get in winter. There is still some level of posterization in the image, which I confess I do not understand. I don't see this with my other cameras, but I don't understand how it can be the camera after stacking 1000 frames.

 

Astrobin full resolution image: https://astrob.in/l7yfk7/0/

 

Edit to add: It seems you have to be logged into Astrobin to see the Full Resolution link. Here is the direct link to the full size image:

 

https://cdn.astrobin...a02a891b8aa.jpg

Attached Thumbnails

  • 02232021_lunar_mosaic_64Mpxl_16bit_gray_compressed_CN.jpg

Edited by Borodog, 24 February 2021 - 01:07 PM.

  • John_Moore, mikerepp, Benschop and 2 others like this

#2 paaul_

paaul_

    Lift Off

  • -----
  • Posts: 5
  • Joined: 16 Feb 2021
  • Loc: France

Posted 24 February 2021 - 11:29 AM

Looks fantastic !

 

Yesterday evening I also recorded some data from the moon using an 8" dob and a DSLR, but I'm having trouble aligning and stacking.

 

Would you mind posting your post-production workflow for this image ? It would really help, since this is exactly what I'm trying to achieve.

 

Thanks in advance.


  • Borodog likes this

#3 aeroman4907

aeroman4907

    Surveyor 1

  • -----
  • Posts: 1,528
  • Joined: 23 Nov 2017
  • Loc: Castle Rock, Colorado

Posted 24 February 2021 - 12:14 PM

Mike,

 

I am running into the same issue with some other posters' images on Astrobin of not being able to view the full size image.  You might want to check with JamesCook1971 to see what he did to fix his.  Here is the thread about this issue: https://www.cloudyni...1-moon-18-2-21/

 

Nevertheless it is looking fairly nice.

 

Steve


  • Borodog likes this

#4 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 01:02 PM

Thank you, paaul.

 

I have found that for stacking DSLR images of the Moon, it is critical to either have excellent tracking, to use flats, or both. If you have an alt-az mount you may also have to deal with field rotation. I stack in Autostakkert and there is an option for field de-rotation, but I don't use it because I have equatorial mounts or platforms, depending on the telescope I am using. If your tracking is off as you shoot, because the sensor size is very large, there can be subtle non-flatness of the frame that is difficult to see visually but can cause stacking problems. If this happens you may have to limit your stack to fewer frames; I don't mean to stack a lower percentage, but rather limit the frames you are stacking to a shorter duration of shooting. The reason to use flats with a DSLR or other large sensor camera is that vignetting can become significant. I am not sure this will adversely affect tracking and stacking, but it can't help and hurts image quality. If you shoot during the daytime, it is easy to shoot "sky flats"; just move the Moon slightly out of frame and shoot the empty sky. Set your shutter speed to make sure all the channels are well exposed, especially the red. Don't touch the ISO. It needs to be the same as the light frames. Autostakkert allows you to generate master calibration frames. I have found that my sky flats seem to work even better as loaded as a master dark rather than a master flat (I have no idea why). Make sure to always use "Surface" rather than "Planet" tracking mode in Autostakkert for the Moon, even if it does not fill the frame (I have definitely seen "Planet" tracking cause alignment/stacking defects). These days I use an alignment point size of 48, regardless of camera, scope, or seeing. It just seems to work the best for me. Do not use rgb alignment for daytime images of the Moon. In fact you probably don't need it at all, so just leave it off.

 

For DSLR (mirrorless, really), my workflow is to use RawTherapee to batch convert all of the raw (Sony .ARW) files to pngs. Then I drag the flat field pngs into Autostakkert and generate a master from them. Then I load that master in AS as either a master flat or master dark (for daytime sky flats only) and drag all of the light frames into AS. Make sure that the option to use a 1.5 pixel horizontal and vertical blur to improve tracking is set. Set Surface mode, leave Noise Robust on 6, and click "Analyze". Once complete, set the alignment point size to 48, Multiscale on. Unless the Moon is full, I always set the minimum brightness to 0 just so nothing is missed in the dark of the terminator. I have played around with moving the alignment region box in the preview pane, but honestly I've never seen much difference, so usually I just leave it in the middle at the default size. Because I can shoot so few frames with the mirrorless camera, I usually go for noise reduction rather than sharpness, especially since I mostly take those images to share on social media where people will rarely view them at full size anyway. Hence I stack most of the frames; I usually shoot about 110-115 lights and stack 100 of them. I usually shoot 40 flats. Don't bother to drizzle, sharpen, RGB align, or any of that in Autostakkert.

 

After stacking I load the raw stack in Astra Image and use the Simple Deconvolution tool, Lucy-Richardson, Gaussian kernel, size usually 1.5 pixels, strength 5, 10 iterations. Once deconvolved, I save, as I have observed Astra Image to hang on subsequent operations after deconvolution. I save to a 16 bit per channel format like png. After that I usually use some combination of gamma/curves correction and/or linear stetch to get the smallest highlights to brilliant white and brighten up the overall image. Sometimes I will raise the black point a tad, sometimes not. These stretching operations are never the same way twice. It depends entirely on how the unstretched image turned out and whatever strikes my fancy at the time.

 

Because there are so few frames stacked, deconvolution will almost certainly introduce noticeable grain. I rarely denoise for a couple of reasons. First, I kind of like the grain; it gives it a more photographic film kind of feel. Second, the denoise algorithm in Astra Image is not very good in my opinion. It introduces more artifacts than it removes. Then again, I always try to denoise after deconvolution, which other members, Tom Glenn and aeroman, have advised me not to do. However it's difficult to see the effect of denoise on the un-sharpened image, so it's difficult to know how much to apply. I need to do more experimentation with this.

 

Finally, for sharing at a lower resolution, I will resize the image and use Unsharp Mask for a final sharpening. Alternately, I may resize the original raw stack first thing and then procede with deconvolution (remember to reduce your kernel size as well), which may result in less noise. Regardless of what the final image looks like I find that posting it to CN significantly degrades it. I can look at my final jpeg and the same file on CN and the CN version looks much worse. I don't know why. I find it very frustrating.

 

Anyway, that is my basic workflow for mirrorless camera lunar images. I do want to emphasis that this particular image was not shot with the mirrorless camera, but rather a cheap obsolete dedicated astro camera, so the capture process was different (SharpCap rather than Sony Remote). For examples of images shot with the mirrorless camera using the workflow above, see one of these threads:

 

https://www.cloudyni...574-02-23-2021/

https://www.cloudyni...-moon-02222021/

https://www.cloudyni...oon-02-21-2021/


  • Benschop, cmas and paaul_ like this

#5 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 01:04 PM

Mike,

 

I am running into the same issue with some other posters' images on Astrobin of not being able to view the full size image.  You might want to check with JamesCook1971 to see what he did to fix his.  Here is the thread about this issue: https://www.cloudyni...1-moon-18-2-21/

 

Nevertheless it is looking fairly nice.

 

Steve

Steve,

 

By any chance are you not logged into Astrobin? I think that may be the issue; only members with logged in may be able to see the Full Resolution button. However, a work around seems to be this, although the full resolution is nothing to write home about because of poor seeing and weird posterization:

 

https://cdn.astrobin...a02a891b8aa.jpg


Edited by Borodog, 24 February 2021 - 01:06 PM.


#6 Tom Glenn

Tom Glenn

    Gemini

  • -----
  • Posts: 3,295
  • Joined: 07 Feb 2018
  • Loc: San Diego, CA

Posted 24 February 2021 - 01:04 PM

I am running into the same issue with some other posters' images on Astrobin of not being able to view the full size image.  You might want to check with JamesCook1971 to see what he did to fix his.  Here is the thread about this issue: https://www.cloudyni...1-moon-18-2-21/

 

Unfortunately, there was really no resolution to the problem in that thread, because the only way we got to see the full sized image was by providing a direct link to the image file itself.  That's not the way it's supposed to work.....somebody should easily be able to access the full sized image simply from the Astrobin main page, by following the links.  Why this problem seems to happen so frequently is confusing, but certainly doesn't make Astrobin appear to be very intuitive or user friendly.  

 

Mike, the image is looking promising, although without the full sized version, I can't comment on any posterization issues, as the only version I can see (1824x1663) doesn't have any obvious problems.  In fact, it looks very nice!  


  • Borodog likes this

#7 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 01:09 PM

Unfortunately, there was really no resolution to the problem in that thread, because the only way we got to see the full sized image was by providing a direct link to the image file itself.  That's not the way it's supposed to work.....somebody should easily be able to access the full sized image simply from the Astrobin main page, by following the links.  Why this problem seems to happen so frequently is confusing, but certainly doesn't make Astrobin appear to be very intuitive or user friendly.  

 

Mike, the image is looking promising, although without the full sized version, I can't comment on any posterization issues, as the only version I can see (1824x1663) doesn't have any obvious problems.  In fact, it looks very nice!  

See the direct link I added to the OP. The Astrobin issue is that you have to be an Astrobin member who is logged in to see the Full Resolution button.



#8 Tom Glenn

Tom Glenn

    Gemini

  • -----
  • Posts: 3,295
  • Joined: 07 Feb 2018
  • Loc: San Diego, CA

Posted 24 February 2021 - 01:24 PM

See the direct link I added to the OP. The Astrobin issue is that you have to be an Astrobin member who is logged in to see the Full Resolution button.

Thanks, the direct link works.  Unfortunately, I don't think the problem with Astrobin is being a member, because I was logged in (I created a membership just for this purpose).  So unless they don't allow free members to view images (which would be a poor way to showcase their product) it must be some other problem.  As Steve has said before, we can see some people's full sized Astrobin images, so something must be happening in the default settings that's not intuitive.  

 

As for your image, I now see what you mean.  When I first hear "posterization" I think of problems with the black level along the terminator, but that's not what's happening here.  This image has severe artifacts throughout the image, as is definitely not the way this should look.  Are all of these problems present in each individual panel?  Not only is there quantization error, but everything is smeared.  How long did each recording take, and is your Dob tracked on an equatorial platform?  If you are tracking in Alt/Az, and if the recording is more than a handful of seconds (I'm not sure of the limit), you will run into problems with stacking lunar images due to field rotation.  Were there any problems with stacking in AS!3, and how many alignment points did you use?


  • Borodog likes this

#9 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 01:54 PM

Thanks, the direct link works.  Unfortunately, I don't think the problem with Astrobin is being a member, because I was logged in (I created a membership just for this purpose).  So unless they don't allow free members to view images (which would be a poor way to showcase their product) it must be some other problem.  As Steve has said before, we can see some people's full sized Astrobin images, so something must be happening in the default settings that's not intuitive.  

 

As for your image, I now see what you mean.  When I first hear "posterization" I think of problems with the black level along the terminator, but that's not what's happening here.  This image has severe artifacts throughout the image, as is definitely not the way this should look.  Are all of these problems present in each individual panel?  Not only is there quantization error, but everything is smeared.  How long did each recording take, and is your Dob tracked on an equatorial platform?  If you are tracking in Alt/Az, and if the recording is more than a handful of seconds (I'm not sure of the limit), you will run into problems with stacking lunar images due to field rotation.  Were there any problems with stacking in AS!3, and how many alignment points did you use?

Strange about the Astrobin thing. I could not see the Full Resolution button for my own image on my workstation until I logged into Astrobin, and then it was available.

 

Each recording was 2000 frames at an average of about 10 fps, so something like 3 & 1/2 minutes. However, it took about 2 hours to shoot all 15 panels because I had to keep moving data off of the internal SSD onto an external SSD, which took about 15 minutes every 2 panels.

 

The Dob is on an Ed O'Brien-built platform and the alignment was bang on. Minimal loss of data at the edge of the frames; the red square in Autostakkert looked the same size it does when used on my worm gear drive C8 on a wedge that has been polar aligned to < 10 arc-seconds in SharpCap Pro.

 

I suspect the smearing comes down to a few things. The first is simply seeing. I wouldn't call it bad, but it was definitely at the bottom end of poor. Coupled with having to limit each panel to only 2000 frames (because of previously discussed drive space and file size issues) and stacking 50% of them to get to a thousand in the stack, that is a recipe for a non very sharp result. I kicked off a re-stack using only 10% of frames just to review if it yields a sharper if noiser result.

 

Second, it has been a while since I checked the collimation of the scope. It may be a touch out, although it clearly isn't terrible. I'll try to find time to check it today before shooting again tonight, but I doubt that is the issue. Third, possibly focus was a bit out as well, as I did not bother with my Bahtinov mask, although like the collimation, if it's out, it is very, very close as I can get it with the stock focuser (counter-intuitively the Dob is easier to focus than the C8, even though the C8 has a finer mechanism). Again, tonight I will not be so lazy and mask it. But I doubt this is the issue.

 

Upon reflection what I think may be significantly contributing to the smearing is the strange posterization in the stack. I think this is potentially harming the ability of Autostakkert to align the points and decide with are sharper. I am going to try doubling the alignment point size and restack a sample panel and see if that improves the smearing.

 

Thinking back, I think you can actually *see* the posterization in the histogram in SharpCap. I'll try to get a screen grab of it tonight. I have an idea that may mitigate it. We'll see.

 

What I do know is that the more time I spend playing with this cheap camera the closer I get to ordering a camera with an IMX183 sensor. :O)


Edited by Borodog, 24 February 2021 - 02:23 PM.


#10 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 02:13 PM

Just ordered a 2 TB internal SSD to replace the 500 GB one. Hopefully no more data worries.



#11 paaul_

paaul_

    Lift Off

  • -----
  • Posts: 5
  • Joined: 16 Feb 2021
  • Loc: France

Posted 24 February 2021 - 02:22 PM

Wow !

 

I didn't expect such a detailed answer.

 

Thank you very much, this is very precious since I was having trouble finding this kind of information online.

 

I'll give it another go now, if I try hard enough it'll work out eventually haha

 

Thanks again and clear skies.


  • Borodog likes this

#12 Tom Glenn

Tom Glenn

    Gemini

  • -----
  • Posts: 3,295
  • Joined: 07 Feb 2018
  • Loc: San Diego, CA

Posted 24 February 2021 - 02:31 PM

Mike, if you are equatorially tracking, then 3.5 minutes is completely fine for a lunar recording.  Several hours spent collecting data will potentially make the mosaic more difficult to blend, but that work is largely done for you by software and you probably would not notice.  It definitely doesn't contribute to the quantization error you are seeing.  Likewise, poor seeing, poor collimation, or poor focus do not cause this problem.  Those cause blurring of details, but not the distinct banding artifacts you are seeing.  I still recommend you try Firecapture, and although I never like to blame the camera (because the camera itself is usually not the problem), if you continue to see this in all of your lunar images then I'm inclined to believe the bargain camera isn't a bargain after all.  


  • Borodog likes this

#13 Tom Glenn

Tom Glenn

    Gemini

  • -----
  • Posts: 3,295
  • Joined: 07 Feb 2018
  • Loc: San Diego, CA

Posted 24 February 2021 - 02:32 PM

And what was the size of your APs and how many did you use?


  • Borodog likes this

#14 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 02:41 PM

Mike, if you are equatorially tracking, then 3.5 minutes is completely fine for a lunar recording.  Several hours spent collecting data will potentially make the mosaic more difficult to blend, but that work is largely done for you by software and you probably would not notice.  It definitely doesn't contribute to the quantization error you are seeing.  Likewise, poor seeing, poor collimation, or poor focus do not cause this problem.  Those cause blurring of details, but not the distinct banding artifacts you are seeing.  I still recommend you try Firecapture, and although I never like to blame the camera (because the camera itself is usually not the problem), if you continue to see this in all of your lunar images then I'm inclined to believe the bargain camera isn't a bargain after all.  

Thanks, Tom. I will definitely try Firecapture. At $90 it is still a fun 8 Mpxl camera to play with, which is really all I am after from it. Nobody on facebook notices the problems we notice here. :O)

 

The posterization is really nothing compared to how truly terrible it looks as a color camera, by the way. I have never resolved the bizarre radial color gradient it produces. Still, I am sure I've already gotten more than $90 worth of enjoyment out of it.



#15 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 02:42 PM

And what was the size of your APs and how many did you use?

48, about 12,000.
 



#16 dcaponeii

dcaponeii

    Viking 1

  • -----
  • Posts: 938
  • Joined: 01 Sep 2019
  • Loc: Waxahachie, TX

Posted 24 February 2021 - 03:15 PM

48, about 12,000.
 

That might be part of the issue.  I usually stack at something closer to 300 AP's.  Sometimes that means using Multi-scale at AP=400 settings.


  • Borodog likes this

#17 Tom Glenn

Tom Glenn

    Gemini

  • -----
  • Posts: 3,295
  • Joined: 07 Feb 2018
  • Loc: San Diego, CA

Posted 24 February 2021 - 03:37 PM

48, about 12,000.
 

Yes, this is potentially very problematic.  Given the resolution of this image, those APs are much too small, and will cause Autostakkert to lock onto false features and noise for alignment, which very likely yields the result you have here.  Before worrying about the mosaic, I would focus on an individual panel and get it looking good.  I would increase the size of the APs to 200 pixels to start with and see what you get.  There's no reason to have more than a few hundred APs per panel when you are troubleshooting.  You can always increase the number later, but only once you have cleared up the fundamental stacking and alignments problem that seems to be occurring.  


  • Borodog likes this

#18 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 03:41 PM

That might be part of the issue.  I usually stack at something closer to 300 AP's.  Sometimes that means using Multi-scale at AP=400 settings.

Perhaps. My thinking is that in general alignment points should be no larger than the minimum required to conclusively register the region from frame to frame. 48 pixels is about 25 times the resolving power and should hence be large enough to ensure registration except in areas of very low gradient, in which case your alignment does not need to be terribly good (and those regions are handled by larger multiscale APs anyway). Perhaps in poor seeing this size should be bumped up.

 

My re-stacks definitely look sharper with only 200 frames stacked, interestingly with apparently little increase in noise. Doesn't help the posterization, though. Those stacks look way better binned 2x2, but then, why use the 8 Mpxl camera instead of the [2] Mpxl camera and skip all of the other limitations?

 

I did discover that the 50% reduced images can tolerate twice the (relative) deconvolution kernel size. i.e. I was deconvolving the full res frames with a 1.5 pixel kernel, and I can still deconvolve the 50% reduced frames with a 1.5 pixel kernel. Which is extremely interesting to me. I knew that reduced images often require re-sharpening at the new, lower resolution, but I had no idea the effect was that large. This requires more testing.


Edited by Borodog, 24 February 2021 - 03:52 PM.


#19 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 03:43 PM

Yes, this is potentially very problematic.  Given the resolution of this image, those APs are much too small, and will cause Autostakkert to lock onto false features and noise for alignment, which very likely yields the result you have here.  Before worrying about the mosaic, I would focus on an individual panel and get it looking good.  I would increase the size of the APs to 200 pixels to start with and see what you get.  There's no reason to have more than a few hundred APs per panel when you are troubleshooting.  You can always increase the number later, but only once you have cleared up the fundamental stacking and alignments problem that seems to be occurring.  

Thanks, Tom. I will give it a shot.



#20 Tom Glenn

Tom Glenn

    Gemini

  • -----
  • Posts: 3,295
  • Joined: 07 Feb 2018
  • Loc: San Diego, CA

Posted 24 February 2021 - 03:54 PM

The idea behind using alignment points at all is that it allows a different subset of frames to be used for regions of the image.  Because seeing induced turbulence can act differentially across a large frame, it makes for a sharper imager if you allow the image to be broken apart.  However, you don't want these AP sizes to be anywhere near the limits of either the telescope or the seeing.  In fact, as an easy test, you can check the box for "global" alignment than "local" in AS!3, which will discard all AP and align entire frames.  This will reduce sharpness, but will completely eliminate any and all problems related to AP selection.  If this eliminates your posterization problem, then you have identified the alignment and stacking as your problem.  If it does not, then you have an underlying problem with the raw data.  


  • Borodog likes this

#21 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 04:06 PM

That's why I picked 48 (as well as it being recommended by another poster a few months back); it seemed like it be sufficient to register the region from frame to frame. Perhaps it isn't. I will say that I seemed to have decent enough results with my other cameras. I don't think this is the source of the posterization; I think rather the posterization may be affecting the alignment. If so then this camera would definitely need larger APs. I'm running with 200 now at your suggestion, with the original 1000 frames stacked so I have an apples to apples comparison of the raw stacks with only 1 think changed at a time. I should end up with:

 

1000 frames stacked, AP = 48

200 frames stacked, AP = 48

1000 frames stacked, AP = 200

200 frames stacked, AP = 200

 

I'll probably run 200 & 1000 frames stacked at AP ~ 100 as well.



#22 aeroman4907

aeroman4907

    Surveyor 1

  • -----
  • Posts: 1,528
  • Joined: 23 Nov 2017
  • Loc: Castle Rock, Colorado

Posted 24 February 2021 - 04:06 PM

Steve,

 

By any chance are you not logged into Astrobin? I think that may be the issue; only members with logged in may be able to see the Full Resolution button. However, a work around seems to be this, although the full resolution is nothing to write home about because of poor seeing and weird posterization:

 

https://cdn.astrobin...a02a891b8aa.jpg

Mike, the link works well and I was able to view full scale.  It is good to see you imaging consistently to hone your imaging and processing skills.

 

My one significant critique is reporting the image to be a 64MP image.  While your native sampling may have matched the presented scale, the image can't reasonably be viewed at that scale, and there is also lots of black space surrounding the Moon.  When I report MP counts on my mosaics, I do it with a very tightly cropped Moon.  A similar crop on your image would be around a 55MP image.  The other factor is that the image needs to maintain a certain level of 'sharpness'.  Your image varies significantly in sharpness (with the area around the Straight Wall being sharper and Mare Nectaris significantly less so).  I understand this is because you took various 'panels' of the Moon with varying seeing and what I suspect is also varying quality of focus.

 

If I were to apply my usual, personal standards for determine proper scale based upon 'sharpness', I would say the Straight Wall looks respectably sharp at 33% of your presented resolution and Mare Nectaris about half that again at 16.5% resolution.  As such, the best resolution is similar to a 6.1MP image and the lowest resolution is closer to a 1.5MP image.  I would typically resize the image to the lowest resolution so that the whole mosaic looks fairly similar in sharpness.

 

I don't want to come off as too harsh in this regard, but I think it is best to present images that are at the right resolution based upon the data.  I have seen many others post similar images claiming huge MP counts when their effective resolution was nowhere near that level, and I am sure many will continue to do so.  I have even been somewhat guilty of this myself in the past by reporting MP counts based upon a 1.5x drizzle.  They look pretty good even at that scale, but I am now shying away from reporting MP counts based upon drizzle data.

 

Despite this critique, I think you will find within this year you'll be able to present images near the MP count you reported here, so keep up the good work!

 

Steve



#23 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 05:23 PM

Not harsh at all, Steve. I always appreciate constructive criticism. I was under no delusions that it was a good capture. :O)

 

So I did some experiments. First there was no improvement due to larger APs . . . but there was no degradation, either. 600 APs instead of 12,000 for the same result is a vast improvement. Second, stacking only 10% vs. 50% noticeably increased sharpness without much increase is the noise. I think this largely comes down to seeing. The seeing just flat out did not support stacking 50% of only 2000 frames. Lastly, down-sampling by 50% made a huge improvement after sharpening. As I said, it allowed me two use twice the relative deconvolution kernal size. Down sampling by 50% increases your signal to noise ratio by 4X. You'd have to shoot 16 times as many frames to get the same increase in SNR. 

 

Regarding what resolution to present an image at, I am just plain unsure. I down-sample images hugely to post here and CN still somehow destroys them compared to the same file viewed locally. I post them and they largely pass without comment, which I take for what it means; they are unremarkable. It's somewhat frustrating. I feel like I have to either pony up thousands of dollars on better equipment or content myself with lower quality images. The one place I think I could get better cheaply is in processing, and I do feel like I'm making some progress there.


  • aeroman4907 likes this

#24 Tom Glenn

Tom Glenn

    Gemini

  • -----
  • Posts: 3,295
  • Joined: 07 Feb 2018
  • Loc: San Diego, CA

Posted 24 February 2021 - 05:57 PM

I post them and they largely pass without comment, which I take for what it means; they are unremarkable. It's somewhat frustrating. 

Nope, don't interpret it that way at all.  The majority of the posts on this forum don't generate intense discussion, and this has been true even for some remarkable posts over the years.  You can often find some hidden gems that have very few "likes" or comments, but which have very good images and/or content.  Reciprocally, depending on the forum in question, "popular" posts do not necessarily represent high quality (although they certainly can).

 

Your imaging is progressing, so don't be frustrated.  The reality is that imaging isn't easy, and there are enormous growing pains along the way.  My recommendation is to step back and focus on the basics.  In this case, my major concern is the quantization error and banding, which doesn't have anything to do with the seeing conditions.  An image can be blurry and lack resolution, yet still not have those artifacts.  So it would be important to determine if the problem is caused by the camera and data capture itself, or with something that is occurring during processing.  What does a cropped region of the raw stack look like (no sharpening, no gamma stretch, just raw)?



#25 Borodog

Borodog

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1,554
  • Joined: 26 Oct 2020

Posted 24 February 2021 - 09:23 PM

Thank you for the encouragement, Tom. I reprocessed this data with a few modifications.

 

First, my earlier comment that a reduced image could tolerate a higher relative deconvolution kernel is simply not true. It may have been try for fully lit panels near the top of the image but it is emphatically NOT true of the high contrast/deep shadow panels at the bottom of the image. The required deconvolution kernel scaled with the image reduction, as expected.

 

Second, I re-stacked everything with 200 pixel APs and 200 frames stacked. After stitching, the first thing I did was reduce the image by 50% (75% in area), and only then proceeded with deconvolution and gamma correction. The resulting image is ~16 Mpxl or however you would like to count it minus the black border. For what it's worth, for viewing the entire disk I prefer a substantial amount of border to set the context (especially for daylight images in the blue sky), but I realize it's unimportant for a full resolution image where you must scroll around the image. In any event, the reduction vastly improved the image in my opinion. Perhaps it needs further reduction, but I am happy with it. For what it's worth I also unsharp masked the final image. It does raise the question of whether it is worth using the 8 Mpxl camera vs the 2 Mpxl camera with a similar sensor size. 

 

The updated image:

 

https://cdn.astrobin...294b8b85ac4.png




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics