Strange RGB pattern?

[Phillyo]

Hi all,

I've just stacked a load of images in Pixinsight using the WeightedBatchPreprocessing script and I have this strange pattern across my image? Camera is the ASI533 which is an RGGB array I believe, but I left the settings to auto (as usual) and I've not seen this before. I calibrated using flats and darkflats at 2 seconds, no bias or darks. It hasn't been an issue before now?

I also stacked about 125 x 90s and 70 x 60s together. Should I stack them seperately the combine?

Thanks for any help! Happy to answer questions if required.

Phil

[ollypenrice]

It's not debayered. I don't know how to ask PI to debayer but presumably you've done it before. There must be a debayer instruction somewhere.

Olly

[Phillyo]

Just now, ollypenrice said:

It's not debayered. I don't know how to ask PI to debayer but presumably you've done it before. There must be a debayer instruction somewhere.

Olly

Hmmm, thanks Olly. I thought I'd selected the debayer. I'll run it again and make sure it's selected.

Phil

[Phillyo]

So I've actually just had a look at the 90sec master light and 60sec master light prior to drizzle integration and neither of them show this pattern, so it's being caused during the drizzle integration part. I'm wondering if drizzling different length subs is the issue? I'm going to drizzle them seperately and report back. Thank you.

[Phillyo]

Update:

Drizzle integration of the 60s subs and 90s subs seperately caused no issues. Redid them combined making sure to debayer and it causes the issue shown. Now I need to figure out why and if there's a way around it!

Phil

[Budgie1]

Have you got "CFA Images" ticked in the Global Options?

[wimvb]

Why do you drizzle? Judging from the provided image,  I don’t think you gain anything. You’re just processing a larger image with more noise. Drizzle is only necessary if the subs are undersampled. In other instances it will give you a noisier image.

[Phillyo]

On 07/11/2020 at 13:30, Budgie1 said:

Have you got "CFA Images" ticked in the Global Options?

Yes. That's part of what is confusing me.

[Phillyo]

22 hours ago, wimvb said:

Why do you drizzle? Judging from the provided image,  I don’t think you gain anything. You’re just processing a larger image with more noise. Drizzle is only necessary if the subs are undersampled. In other instances it will give you a noisier image.

I'm imaging at a scale of around 6"/pixel so I'm severely undersampling. The image I linked is after drizzle. Before drizzle is very noticeably worse

[wimvb]

If the combination of 60 and 90 s exposures causes this, then just drizzle integrate each separately. Then combine using pixelmath.

(integration _60 + integration_90)/2

(Or something similar)

Edited November 8, 2020 by wimvb

[Phillyo]

Yeah I did a HDRcombination and that worked. I was just a little surprised and confused that the drizzle would cause that strange pattern.

[vlaiv]

1 hour ago, Phillyo said:

I'm imaging at a scale of around 6"/pixel so I'm severely undersampling. The image I linked is after drizzle. Before drizzle is very noticeably worse

Fact that you are at 6"/pixel does not mean you are under sampled.

This would probably be true for diffraction limited scope but it won't be true for lens - and I suppose you are referring to Samyang 135mm F/2 lens here?

Most lenses are not diffraction limited (even if they are deemed very sharp by photographic standards).

Not sure if you've seen these two graphs - they are from Samyang website. Look at 30 lines per mm (grey lines - around 90% MTF) . This is effectively proper sampling for pixel size of 16.666µm. You have pixel size of 3.76µm - which is x4.4 smaller.

I recently measured sharpness of Samyang 85mm F/1.4 using artificial star at center of the field and F/2.8 aperture and got results that say that at 4.8µm pixel size I'm still oversampling by almost factor of two.

 

 

[Phillyo]

@vlaiv I'm not even going to pretend I understand much of what that means, however I'll take your word for it! That being said, before and after doing 2xdrizzle on my images makes a big difference. 

Images show before and after drizzle of roughly the same area of sky around M31.

[vlaiv]

4 minutes ago, Phillyo said:

Images show before and after drizzle of roughly the same area of sky around M31.

That is on 300% zoom in PI I presume?

You'll see "blocky" stars in such display because it uses only nearest neighbor interpolation when zooming (so you can see individual pixels). Take same undrizzled image and scale it to be the same size as drizzled image with more sophisticated algorithm like Lanczos or Cubic B-Spline or similar and then compare the results.

[Phillyo]

7 hours ago, vlaiv said:

That is on 300% zoom in PI I presume?

You'll see "blocky" stars in such display because it uses only nearest neighbor interpolation when zooming (so you can see individual pixels). Take same undrizzled image and scale it to be the same size as drizzled image with more sophisticated algorithm like Lanczos or Cubic B-Spline or similar and then compare the results.

I'm not sure how to do that but I'll give it a go tomorrow to see if I can figure it out. If I don't need to drizzle the data then I won't. It'll save me a lot of time in the preprocessing phase!!

Thanks,

Phil.

[wimvb]

43 minutes ago, Phillyo said:

I'm not sure how to do that but I'll give it a go tomorrow to see if I can figure it out.

(Integer) resample in Pixinsight should do it

[Adam J]

On 07/11/2020 at 10:39, Phillyo said:

Update:

Drizzle integration of the 60s subs and 90s subs seperately caused no issues. Redid them combined making sure to debayer and it causes the issue shown. Now I need to figure out why and if there's a way around it!

Phil

Are you using the correct pattern. 

[vlaiv]

50 minutes ago, Phillyo said:

I'm not sure how to do that but I'll give it a go tomorrow to see if I can figure it out. If I don't need to drizzle the data then I won't. It'll save me a lot of time in the preprocessing phase!!

Thanks,

Phil.

https://pixinsight.com/doc/tools/Resample/Resample.html

Don't use integer resample - that is for software binning.

Here is an example for you (done in ImageJ):

vs

First one is just "zoomed in" at 300% image that is close to being under sampled (nearest neighbor - pixels are visible) . Below is exactly the same image but up scaled 300% using proper scaling method - no more pixels.

 

[Phillyo]

1 hour ago, Adam J said:

Are you using the correct pattern. 

Yes, RGGB used for all 3 times I tried it (60s alone, 90s alone and mixed 60/90s subs)

[Phillyo]

So I tried to resample it last night using different scaling methods and none of the results were as clean as the 2xdrizzle. Much more noise and much harder to process. I'm probably going something wrong but I'll stick with the 2xdrizzle for now.