Strange artifacts in image viewer when switching from Calibrated to Registered image options

I was about to create a new topic about this until I found your post. I'm facing the exact same problem and I'm not sure if it affects the final image or not. It doesn't look right though.

Me too! Doesn't seem to be affecting integration results, but who know, perhaps it does in some minor way. Probably just the viewer though.

Hi all @mestutters , @msid, @jm

Thank you very much for reporting this ! This in fact is not an issue, and also not a bug 🙂 !

It actually is drizzle working and showing you where drizzle drops are falling in the integration target field of view and scale. Zoom in on pixel scale and you will see that some pixels will get drizzle drops, others not... it is in principal how drizzle works.

If you have drizzle or Bayer/X-Trans drizzle enabled in 6) integrate at the bottom of the menu, the registered view shows the drizzle droplets. Drizzle reconstruction will, depending on the integration scale and the drizzle kernel and droplet size, not flll in all pixels in the target grid for integration and the pixels where droplets are falling, are falling there with integration weights as well.

Depending on the registration parameters( that will move, rotate, scale, undistort your images to fit to the reference) this will lead to these weird mathematical structures in these registered views. I personally think it is awesome that APP is showing you how drizzle actually works like this. This has been in APP already for a long time.

As you have experienced, this does not affect the integration, the results are as expected. You can also check/test the following way, look at such a registered view, then change the droplet size and reload the registered view, you will see it looks different because of different drizzle parameters. You can also test the following, disable drizzle and use interpolation in 6) integrate and reload the registered view, you will see it looks normal without the drizzle "artefacts" 😉

Mabula

Thanks Mabula, it always amazes me just how sophisticated APP is, so much control/detail.

To this point, this is a perfect example of how badly we need manual for APP. There is just so much (too much?) to remember/know, it is not possible for someone to keep track of it all. I know one is "coming with V2" but I think I may "pass away" before I see it (joking of course) 😉 .

Let's hope it's very soon!

Thanks again

Hi Mabula,

Thanks for the explanation.  For my own imaging rig I don't think that drizzling provides any benefit but using e.g. the Russ Croman MTF Analyser tool, suggests that using drizzle with certain iTelescope rigs should be capable of extracting some additional fine detail, hence my interest.

Now I appreciate what APP is doing I can see that this feature is potentially most useful?  I have just tried to reproduce the results I obtained earlier but am not seeing the same effects.  I am guessing this is because I've since obtained more data for M31.  Is this correct reasoning?  Assuming so, then I guess that for any user seeing this type of result when using drizzle (or Bayer drizzle) then the message is to acquire more data and/or to reduce the  scale factor?   I assume if I reduce the input image count then I should start to see my earlier results being reproduced?

With reference to my original screen shots, would I be correct to assume that the darkest areas are the parts of the image receiving the fewest drizzle droplets or is there something else going on with respect to the shading/colouring?

Do you have any information about the implications of changing drop size?   I've looked at the orignal papers describing drizzle but see nothing obvious about how to select an optimum size? Does a suitable drop size have any relationship to scaling factor?  You've mentioned elsewhere I recall that Scale 2 with Drop size 0.5 would be equivalent to 'traditional' drizzling?   And does drop size relate to: the input grid, the target grid, original pixel size, something else?    I notice that drop size can be greater or less than 1.   What effects might be observed by transitioning drop size from greater than to less than 1?

Regards

Mike

@mabula-admin 

This is awesome, thanks for clearing that up!

Is there any way we can use this "special view" in order to find the optimal scale and droplet values when drizzling?

I'm pretty much always drizzling at 250mm focal length and 3.76um pixel size: I set the scale to 3.0 and then create multiple CROP integrations using different droplet values (0.9, 0.8, 0.7, ... 0.1). Finally I  check the SNR and NOISE values found in the metadata of the final stacks in order to decide which scale/droplet combination would give me the best result.

This gets the job done but is very time consuming, especially when integrating thousands of images. Is there a "better" way to do this?

PS: I don't know if it is a coincidence, but scale 3.0 and droplet size 0.1 gives me the best results. 

Posted by: @jm

↑

Thanks Mabula, it always amazes me just how sophisticated APP is, so much control/detail.

To this point, this is a perfect example of how badly we need manual for APP. There is just so much (too much?) to remember/know, it is not possible for someone to keep track of it all. I know one is "coming with V2" but I think I may "pass away" before I see it (joking of course) 😉 .

Let's hope it's very soon!

Thanks again

🙂 you are most welcome @jm

Working on it !

Posted by: @mestutters

↑

Hi Mabula,

Thanks for the explanation.  For my own imaging rig I don't think that drizzling provides any benefit but using e.g. the Russ Croman MTF Analyser tool, suggests that using drizzle with certain iTelescope rigs should be capable of extracting some additional fine detail, hence my interest.

Now I appreciate what APP is doing I can see that this feature is potentially most useful?  I have just tried to reproduce the results I obtained earlier but am not seeing the same effects.  I am guessing this is because I've since obtained more data for M31.  Is this correct reasoning?  Assuming so, then I guess that for any user seeing this type of result when using drizzle (or Bayer drizzle) then the message is to acquire more data and/or to reduce the  scale factor?   I assume if I reduce the input image count then I should start to see my earlier results being reproduced?

With reference to my original screen shots, would I be correct to assume that the darkest areas are the parts of the image receiving the fewest drizzle droplets or is there something else going on with respect to the shading/colouring?

Do you have any information about the implications of changing drop size?   I've looked at the orignal papers describing drizzle but see nothing obvious about how to select an optimum size? Does a suitable drop size have any relationship to scaling factor?  You've mentioned elsewhere I recall that Scale 2 with Drop size 0.5 would be equivalent to 'traditional' drizzling?   And does drop size relate to: the input grid, the target grid, original pixel size, something else?    I notice that drop size can be greater or less than 1.   What effects might be observed by transitioning drop size from greater than to less than 1?

Regards

Mike

 

Hi @mestutters,

"Now I appreciate what APP is doing I can see that this feature is potentially most useful? I have just tried to reproduce the results I obtained earlier but am not seeing the same effects. I am guessing this is because I've since obtained more data for M31. Is this correct reasoning? Assuming so, then I guess that for any user seeing this type of result when using drizzle (or Bayer drizzle) then the message is to acquire more data and/or to reduce the scale factor? I assume if I reduce the input image count then I should start to see my earlier results being reproduced?"

Drizzle with the purpose of getting some additional fine detail requires 3 things:

1) enough data, usually means a lot of subs !

2) undersampled data

3) well dithered data

Now, the images that show drizzle working with the registered view of a single sub, simply show how that particular image adds to the drizzled integration. If you supply enough data that is well dithered and undersampled. Those drizzle artefacts will be evened out in the final integraton and thus the integration will look good. If you have too little data of perhaps too small droplet size,  you will start to see drizzle artefacts in the integration because of that. This should also explain why drizzle is a big ! noise injector and thus requires more data than regular data interpolation for integration of the subs.

"With reference to my original screen shots, would I be correct to assume that the darkest areas are the parts of the image receiving the fewest drizzle droplets or is there something else going on with respect to the shading/colouring?"

No, that is correct, in those darker areas, more pixels are not receiving drizzle droplets for that particular image adding to the integration. Another one of your subs might compensate in that particular area though 😉

"Do you have any information about the implications of changing drop size?   I've looked at the orignal papers describing drizzle but see nothing obvious about how to select an optimum size? Does a suitable drop size have any relationship to scaling factor?  You've mentioned elsewhere I recall that Scale 2 with Drop size 0.5 would be equivalent to 'traditional' drizzling?   And does drop size relate to: the input grid, the target grid, original pixel size, something else?    I notice that drop size can be greater or less than 1.   What effects might be observed by transitioning drop size from greater than to less than 1?"

First of all, there is no such thing as optimum droplet size. It depends on the data itself, the scale factor you choose, the actual drizzle kernel  and on your subjective approval of noise versus sharpness in the final result. The lower the droplet size, the sharper the result will be (if data allows it in terms of undersampling) and the noiser it will be. How much noise in the integration you find acceptable seems to be a very subjective thing as you would probably agree.

Please check these 2 older posts on our forum related to these questions 😉 :

https://www.astropixelprocessor.com/community/tutorials-workflows/drizzle-for-mono-cameras/#post-4257

https://www.astropixelprocessor.com/community/tutorials-workflows/bayer-drizzle/#post-15814

Posted by: @msid

↑

@mabula-admin 

This is awesome, thanks for clearing that up!

Is there any way we can use this "special view" in order to find the optimal scale and droplet values when drizzling?

I'm pretty much always drizzling at 250mm focal length and 3.76um pixel size: I set the scale to 3.0 and then create multiple CROP integrations using different droplet values (0.9, 0.8, 0.7, ... 0.1). Finally I  check the SNR and NOISE values found in the metadata of the final stacks in order to decide which scale/droplet combination would give me the best result.

This gets the job done but is very time consuming, especially when integrating thousands of images. Is there a "better" way to do this?

PS: I don't know if it is a coincidence, but scale 3.0 and droplet size 0.1 gives me the best results. 

Hi @msid,

You are most welcome !

No, not really, because the registered view of single image will not tell enough about how the integration will look in the end. It depends completely on the amount of data, the amount of undersampling and the amount of dithering between the subs... So the best answer I can give is that your method is the way to go combined with building experience with datasets and drizzling and experiencing what is possible 😉

But, scale 3.0 and droplets of 0,1 is very extreme I must point out. That should lead to a lot of noise and probably only little improvement in gained sharpness when compared to using scale 2.0 and droplets of 0,5 - 0.7. In my experience, you need very undersampled data to have benefit of droplets smaller than 0.5. Do you see clear improvements with so small droplet size? 250mm focal length is rather small so It could be the case if you shoot under great sky conditions as well?

Mabula