NGC4565 - Needle galaxy

[EyeGuy]

I found a new spot in the garden, more shielded from street lights, so this was taken with no filters a couple of nights ago, under bortle 6 skies.

5 min subs X 32 = 160 min integration time

Processed in PI and Photoshop.

Barry

[MarkAR]

Nice one, got a few extra fuzzies as well.

[Adam J]

8 hours ago, EyeGuy said:

I found a new spot in the garden, more shielded from street lights, so this was taken with no filters a couple of nights ago, under bortle 6 skies.

5 min subs X 32 = 160 min integration time

Processed in PI and Photoshop.

Barry

Which scope are you using?

[EyeGuy]

It was a TEC 180 FL - I tagged it at the top of the thread but the tags aren't very conspicuous on SGL.

Barry

[vlaiv]

Nice capture.

I see that you drizzled - according to title of the image. Why is that?

[EyeGuy]

Well spotted!

It's funny - I didn't think drizzle would make any difference at all because the sampling is about right with 5.34um pixels, but it definitely seemed to increase the detail a tiny bit when I zoomed in, and it made edges more pleasing.

Barry

[oymd]

very very nice image..

[EyeGuy]

1 minute ago, oymd said:

very very nice image..

Thanks! Glad you like it. I'm really starting to like the EOS Ra.

[vlaiv]

2 minutes ago, EyeGuy said:

Well spotted!

It's funny - I didn't think drizzle would make any difference at all because the sampling is about right with 5.34um pixels, but it definitely seemed to increase the detail a tiny bit when I zoomed in, and it made edges more pleasing.

Barry

You don't happen to have a comparison? I wonder because I maintain that drizzling does not improve anything - just hurts people's data / end result.

[oymd]

9 minutes ago, vlaiv said:

Nice capture.

I see that you drizzled - according to title of the image. Why is that?

HI Vlad

How did you know Eyeguy drizzled? What am I missing?

[EyeGuy]

Just now, vlaiv said:

You don't happen to have a comparison? I wonder because I maintain that drizzling does not improve anything - just hurts people's data / end result.

I don't, I'm afraid, and it would be v hard to show on the website.

I was of the same opinion until this image, and it's not like the difference was striking, but I was convinced it was a little better with drizzle when I examined the 2 with same zoom settings in PI. Maybe just my imagination as the sampling really is just about spot on for my scope.

Barry

[EyeGuy]

5 minutes ago, oymd said:

HI Vlad

How did you know Eyeguy drizzled? What am I missing?

It's in the filename. When I'm processing I name the file according to what I've done at each step so I can retrace. Vlad clearly has eagle eyes. 🙂

[vlaiv]

4 minutes ago, oymd said:

HI Vlad

How did you know Eyeguy drizzled? What am I missing?

I like to examine images in full - that involves right click and open image in new Tab on my Firefox browser. There you can see actual title of the image and zoom to 100% or 1:1.

Title of this image is

NGC4565_drizzle_integration_crop_DBE_curves_PMCC_crop.jpg

That kind of suggests what was done to it - among other things, drizzle was applied.

3 minutes ago, EyeGuy said:

I don't, I'm afraid, and it would be v hard to show on the website.

I was of the same opinion until this image, and it's not like the difference was striking, but I was convinced it was a little better with drizzle when I examined the 2 with same zoom settings in PI. Maybe just my imagination as the sampling really is just about spot on for my scope.

Barry

Did you by any chance do Bayer drizzle - as that is something that would work. I'm not sure if PI has that algorithm. It uses drizzle instead of interpolation to debayer image and since it won't try to reconstruct detail smaller than a single pixel - just fill in missing data - it actually works.

[EyeGuy]

Just now, vlaiv said:

I like to examine images in full - that involves right click and open image in new Tab on my Firefox browser. There you can see actual title of the image and zoom to 100% or 1:1.

Title of this image is

NGC4565_drizzle_integration_crop_DBE_curves_PMCC_crop.jpg

That kind of suggests what was done to it - among other things, drizzle was applied.

Did you by any chance do Bayer drizzle - as that is something that would work. I'm not sure if PI has that algorithm. It uses drizzle instead of interpolation to debayer image and since it won't try to reconstruct detail smaller than a single pixel - just fill in missing data - it actually works.

I tried Bayer drizzle - but PI told me it wasn't a Bayer image, though it was, so I gave up quickly and just went with the usual drizzle. If I get a chance I'll process it without the drizzle and post here for comparison - maybe a very small closeup would be useful.

You're probably right that it made no difference - but it sure makes enormous files.

[oymd]

4 minutes ago, EyeGuy said:

It's in the filename. When I'm processing I name the file according to what I've done at each step so I can retrace. Vlad clearly has eagle eyes. 🙂

yes...agreed...

Vlad is on another level!

 

[vlaiv]

15 minutes ago, EyeGuy said:

I tried Bayer drizzle - but PI told me it wasn't a Bayer image, though it was, so I gave up quickly and just went with the usual drizzle. If I get a chance I'll process it without the drizzle and post here for comparison - maybe a very small closeup would be useful.

You're probably right that it made no difference - but it sure makes enormous files.

If you want to use Bayer drizzle - you should not debayer your subs first. Not sure what your workflow is, but Bayer drizzle requires mono raw calibrated subs (from OSC camera obviously).

[alan potts]

That is a very nice image,

Alan

[Adam1234]

That's lovely

[EyeGuy]

6 hours ago, vlaiv said:

If you want to use Bayer drizzle - you should not debayer your subs first. Not sure what your workflow is, but Bayer drizzle requires mono raw calibrated subs (from OSC camera obviously).

Thanks for that - in this instance the camera was a Canon EOS Ra. My typical workflow for preprocessing is Adam Bloch's method. It's basically the same as the script, but done manually to look for trouble as you go.  So the debayering happens, I think, as PI loads the images. How would I go about processing the greyscale images and then combining? I'm afraid to mess with the camera raw loading options as the last time I did that I screwed everything up 🙂

Barry

[vlaiv]

48 minutes ago, EyeGuy said:

Thanks for that - in this instance the camera was a Canon EOS Ra. My typical workflow for preprocessing is Adam Bloch's method. It's basically the same as the script, but done manually to look for trouble as you go.  So the debayering happens, I think, as PI loads the images. How would I go about processing the greyscale images and then combining? I'm afraid to mess with the camera raw loading options as the last time I did that I screwed everything up 🙂

Barry

Here is what general workflow should look like - but I don't use PI so I can't give specific details on how to do each step:

1. Load subs raw subs and convert them to 32bit mono image

2. Load calibration subs (darks, flats, flat darks)

3. do calibration as you would for mono image. Bare in mind that your calibration subs should also be mono. This means - created in the same way - load raw subs and treat them as mono when making masters.

4. Use such calibrated but undebayered subs in bayer drizzle stacking method.

Regular drizzle works by taking each pixel and then "reducing" its size - thus creating "empty" space around it. It then aligns such pixels to output image and places pixels on it (drizzle pixels over output image). This is in principle what any resampling algorithm will do when you change resolution - but it reduces pixels even more - to a single point (in math terms) - it just does not drizzle such points on output image as that would be pointless (pun?) since points have no size. It works in opposite direction - it calculates expected value between point samples by applying reverse transform from output image (takes coordinates of output image pixel and calculates where it should lie on original image).

In any case - regular drizzle won't work, or it will produce less SNR than resampled integration explained above. However, bayer drizzle will work - since pixels are already smaller and you don't need to add artificial space between pixels - it has it already - but the thing is - bayer drizzle won't produce larger image. It will produce the same pixel count image as regular debayering methods - only marginally sharper (if there is undersampling with bayer matrix in the first place).

[EyeGuy]

1 hour ago, vlaiv said:

Here is what general workflow should look like - but I don't use PI so I can't give specific details on how to do each step:

1. Load subs raw subs and convert them to 32bit mono image

2. Load calibration subs (darks, flats, flat darks)

3. do calibration as you would for mono image. Bare in mind that your calibration subs should also be mono. This means - created in the same way - load raw subs and treat them as mono when making masters.

4. Use such calibrated but undebayered subs in bayer drizzle stacking method.

Regular drizzle works by taking each pixel and then "reducing" its size - thus creating "empty" space around it. It then aligns such pixels to output image and places pixels on it (drizzle pixels over output image). This is in principle what any resampling algorithm will do when you change resolution - but it reduces pixels even more - to a single point (in math terms) - it just does not drizzle such points on output image as that would be pointless (pun?) since points have no size. It works in opposite direction - it calculates expected value between point samples by applying reverse transform from output image (takes coordinates of output image pixel and calculates where it should lie on original image).

In any case - regular drizzle won't work, or it will produce less SNR than resampled integration explained above. However, bayer drizzle will work - since pixels are already smaller and you don't need to add artificial space between pixels - it has it already - but the thing is - bayer drizzle won't produce larger image. It will produce the same pixel count image as regular debayering methods - only marginally sharper (if there is undersampling with bayer matrix in the first place).

Thanks for taking the time to give that very detailed description, Vlad. It's very much appreciated. 

I remember trying to process some Nikon NEF raw files as individual mono, but I screwed the whole thing up and it ended up a mess. I could try to give it a go again with what you've said above. I don't think I'd bother on the data from the Needle - especially since it's perfectly sampled for my setup, but I have a Moravian OSC which I kind of bought in error. I read the 6um pixel size and thought that sounded great - but pixel size on normal commercial cameras is very different to pixel size on astronomical cameras. When you figure the bayer matrix in the real pixel size is 12um - which is badly undersampled. (at least, I think this is how it works - anyone can correct me if I'm wrong).

Anyway, I'll give that a go like you suggest - but I think I'm coming down with Covid-19 now so we'll have to see...

Thanks again, and best wishes,

Barry

[vlaiv]

2 minutes ago, EyeGuy said:

but I think I'm coming down with Covid-19 now so we'll have to see...

Hope you'll be ok!

2 minutes ago, EyeGuy said:

I remember trying to process some Nikon NEF raw files as individual mono, but I screwed the whole thing up and it ended up a mess. I could try to give it a go again with what you've said above. I don't think I'd bother on the data from the Needle - especially since it's perfectly sampled for my setup, but I have a Moravian OSC which I kind of bought in error. I read the 6um pixel size and thought that sounded great - but pixel size on normal commercial cameras is very different to pixel size on astronomical cameras. When you figure the bayer matrix in the real pixel size is 12um - which is badly undersampled. (at least, I think this is how it works - anyone can correct me if I'm wrong).

All OSC sensors behave the same - they have bayer matrix and are in principle sampling at twice the lower rate than pixel size would suggest. This is not something that is exclusive in dedicated astronomy cameras.

Although single sub will sample at rate that corresponds to double pixel size - stack won't quite behave like that. In fact that depends on what sort of debayering one applies. Probably best way to do it would be Bayer drizzle algorithm or upsampled integration.

I don't think that later is available as "well known algorithm" - but you could easily replicate it for testing purposes. As long as you can do split debayering (which does not use interpolation nor super pixel mode, but instead produces smaller images for each channel - and twice as many subs for green because there are two green pixels for each red and blue) - you can resample your images to larger size using lanczos 3 resampling and then integrate those - that will be closest thing to upsampled integration. Of course - bayer drizzle is alternative to be tried.