Help Me Understand The End Game Benefit of Drizzle - Forums

Topic starter

Mar 8, 2025

For almost a year now, maybe longer, I have been dutifully dithering my data at least every 2 subs and by 5 pixels. I have done this because…well becuase you're supposed to according to everyone. I know there are benefits beyond drizzling and it is good practice. The first few times I used the dithered data and applied a 2X drizzle in PixInsight I ended up, as expected, with a "normal" Master Light and the gigantic 2X Drizzled Master Light along with the other normal output of WBPP. Okay fair enough. I kept on gathering all my data dithered but never apply a drizzle because I really never saw a benefit other than a giant file size that choked my computer.

In those early attempts I was working from a Dell Laptop that wasn't up to the task. Now I have a Desktop PC that is fully capable and can work with the huge files and thanks to the GPU BlurXterminator doesn't take hours to run on an 800MB file. But I guess I am still missing the point. Yes I have an enormous file and the detail is there but what then? How do I use this huge file to my advantage? I am assuming it has to do with down sampling but I am clueless about that. It would be great if someone could explain at the end of the day what the benefit of a drizzled file is and how and where to exploit it? Apologize if this is a dumb question.

Engaging

andrea tasselli

Frequent contributor

Mar 8, 2025 · Edited: Mar 8, 2025

There is only one ultimate benefit of drizzling, assuming all the attendant conditions are met, that is to recover some lost resolution if you're significantly under-sampling your FWHM. You pay a heavy price in SNR, however. So ultimately it might not worth the price. And BXT might not play nice with drizzled files, as I have discovered to my chagrin.

BTW, I never ever ever ever dithered my subs. Like, ever.

Engaging

D. Jung

Mar 8, 2025

As @andrea tasselli said, you can use drizzle to recover resolution. Here is a good explanation of the process https://www.stsci.edu/~fruchter/dither/drizzle.html

Another point is that dithering helps to remove fixed pattern noise (fpn) like dead pixels (by making sure that the image is shifted across the sensor while the fixed pattern noise of the sensor doesn't change).

Concise

Tony Gondola

Frequent contributor

Mar 8, 2025

I drizzle to prevent walking noise, the result is pretty ugly if I don't.

Tony Gondola

Frequent contributor

Mar 8, 2025

Tony Gondola:
I dither to prevent walking noise, the result is pretty ugly if I don't.

correction above...

Klaus Haaken

Mar 8, 2025

Tony Gondola:
Tony Gondola:
I dither to prevent walking noise, the result is pretty ugly if I don't.

correction above...

The same reason why I also dither.

GiffS

Topic starter

Mar 8, 2025

So the drizzle benefit only really comes with under sampled data. Dithering is something I have been and will continue to do for reasons mentioned. It used to be costly time wise when I was using an AVX because it would take a long time to settle. Since I got the AM5 that is no longer a big issue.

Thanks for the replies.

Kay Ogetay

Mar 8, 2025 · Edited: Mar 8, 2025

People quite often ask me how I get such results with Redcat 51, what is my secret. The thing is there is no secret. Understanding what Drizzling does is the key.

I have a significant undersampling with Redcat 51 and 2600MM. For example, this is the full image I get:

But when I apply x3 drizzling to this image, I'm able to get this image:

This result is pretty close to what one (on average) gets with much higher focal lengths (compared to 250mm). This is the power of drizzling in undersampled cases. But if you ask around, many will say 2x drizzling is unnecessary. Let alone 3x... Well, the result is right here.

People with different setups that don't benefit from drizzling much try and see no difference and blame the drizzling. Understanding what the process does is quite important to avoid such confusions.

Helpful Insightful Respectful Engaging

Tony Gondola

Frequent contributor

Mar 8, 2025

Kay Ogetay:
People quite often ask me how I get such results with Redcat 51, what is my secret. The thing is there is no secret. Understanding what Drizzling does is the key.

I have a significant undersampling with Redcat 51 and 2600MM. For example, this is the full image I get:

But when I apply x3 drizzling to this image, I'm able to get this image:

This result is pretty close to what one (on average) gets with much higher focal lengths (compared to 250mm). This is the power of drizzling in undersampled cases. But if you ask around, many will say 2x drizzling is unnecessary. Let alone 3x... Well, the result is right here.

People with different setups that don't benefit from drizzling much try and see no difference and blame the drizzling. Understanding what the process does is quite important to avoid such confusions.

Can you present both images at the same image scale? I think that would show better the point you're trying to make.

Well Written Respectful Concise

Jeffery Richards

Mar 9, 2025

Tony Gondola:
Kay Ogetay:
People quite often ask me how I get such results with Redcat 51, what is my secret. The thing is there is no secret. Understanding what Drizzling does is the key.

I have a significant undersampling with Redcat 51 and 2600MM. For example, this is the full image I get:

But when I apply x3 drizzling to this image, I'm able to get this image:

This result is pretty close to what one (on average) gets with much higher focal lengths (compared to 250mm). This is the power of drizzling in undersampled cases. But if you ask around, many will say 2x drizzling is unnecessary. Let alone 3x... Well, the result is right here.

People with different setups that don't benefit from drizzling much try and see no difference and blame the drizzling. Understanding what the process does is quite important to avoid such confusions.

Can you present both images at the same image scale? I think that would show better the point you're trying to make.

Not wanting to put words in Kay's mouth, but I suspect it's something along the lines of:

You will be left with two sets of masters. For the full FOV image just use the basic master but the drizzle master will allow you to crop an item of interest in the full FOV (the Crescent in this case) and have a very good high resolution image of a fractional part of the overall FOV. Two presentations from a single imaging session. Or maybe I'm completely off base since I'm not all that up on drizzling.

Respectful Supportive

GiffS

Topic starter

Mar 9, 2025

Jeffery Richards:
Tony Gondola:
Kay Ogetay:
People quite often ask me how I get such results with Redcat 51, what is my secret. The thing is there is no secret. Understanding what Drizzling does is the key.

I have a significant undersampling with Redcat 51 and 2600MM. For example, this is the full image I get:

But when I apply x3 drizzling to this image, I'm able to get this image:

This result is pretty close to what one (on average) gets with much higher focal lengths (compared to 250mm). This is the power of drizzling in undersampled cases. But if you ask around, many will say 2x drizzling is unnecessary. Let alone 3x... Well, the result is right here.

People with different setups that don't benefit from drizzling much try and see no difference and blame the drizzling. Understanding what the process does is quite important to avoid such confusions.

Can you present both images at the same image scale? I think that would show better the point you're trying to make.

Not wanting to put words in Kay's mouth, but I suspect it's something along the lines of:

You will be left with two sets of masters. For the full FOV image just use the basic master but the drizzle master will allow you to crop an item of interest in the full FOV (the Crescent in this case) and have a very good high resolution image of a fractional part of the overall FOV. Two presentations from a single imaging session. Or maybe I'm completely off base since I'm not all that up on drizzling.

Now that makes perfect sense to me. Thank you.

Tony Gondola

Frequent contributor

Mar 9, 2025

If I measured it right, the average FWHM of the image is under 1" which is well below the resolution limit of a 51mm aperture (2.28") Apparently it can be a very powerful technique indeed.

Arun H

Frequent contributor

Mar 9, 2025 · Edited: Mar 9, 2025

Tony Gondola:
If I measured it right, the average FWHM of the image is under 1" which is well below the resolution limit of a 51mm aperture (2.28") Apparently it can be a very powerful technique indeed.

Drizzle cannot exceed the theoretical resolution from convolving the effect of aperture and the atmosphere. Since some resolution is always lost to the seeing, the maximum obtained resolution will be slightly less than what may be predicted based on the Rayleigh limit. What drizzle does is recover resolution lost if the pixel density is less than what is required by the Nyquist criterion.

For 51mm aperture and perfect seeing, you get a resolution of 2.71 arcseconds. With a 3.76 micron pixel size and 250mm focal length, you are at 3.1"/pixel whereas you need to be at roughly 1.35", that is, if the seeing is perfect, you are under sampled. Drizzling, therefore, can, assuming good seeing conditions, get you closer to the theoretical limit of 2.71".

CFA drizzle for Bayer array sensors uses, I think, a very similar principle. With a Bayer array, you will almost always be heavily undersampled for each color. Assuming you dithered and have a sufficient number of subs, you can reconstruct a higher resolution image for each color instead of using Bayer interpolation.

Well Written Helpful Insightful

Tony Gondola

Frequent contributor

Mar 9, 2025

Arun H:
Tony Gondola:
If I measured it right, the average FWHM of the image is under 1" which is well below the resolution limit of a 51mm aperture (2.28") Apparently it can be a very powerful technique indeed.

Drizzle cannot exceed the theoretical resolution from convolving the effect of aperture and the atmosphere. Since some resolution is always lost to the seeing, the maximum obtained resolution will be slightly less than what may be predicted based on the Rayleigh limit. What drizzle does is recover resolution lost if the pixel density is less than what is required by the Nyquist criterion.

For 51mm aperture and perfect seeing, you get a resolution of 2.71 arcseconds. With a 3.76 micron pixel size and 250mm focal length, you are at 3.1"/pixel whereas you need to be at roughly 1.35", that is, if the seeing is perfect, you are under sampled. Drizzling, therefore, can, assuming good seeing conditions, get you closer to the theoretical limit of 2.71".

CFA drizzle for Bayer array sensors uses, I think, a very similar principle. With a Bayer array, you will almost always be heavily undersampled for each color. Assuming you dithered and have a sufficient number of subs, you can reconstruct a higher resolution image for each color instead of using Bayer interpolation.

That's what I had assumed. My question wasn't about the limits of what drizzling can do. It was how can you get an image with a final measured average FWHM of under 1.0" with a 2 inch aperture? I must be measuring things incorrectly.

Well Written

Kay Ogetay

Mar 9, 2025

Jeffery Richards:
Tony Gondola:
Tony Gondola (Gondola)

Not wanting to put words in Kay's mouth, but I suspect it's something along the lines of:

You will be left with two sets of masters. For the full FOV image just use the basic master but the drizzle master will allow you to crop an item of interest in the full FOV (the Crescent in this case) and have a very good high resolution image of a fractional part of the overall FOV. Two presentations from a single imaging session. Or maybe I'm completely off base since I'm not all that up on drizzling.

Yes, pretty much what I wanted to say. It's nice because I can shoot multiple targets at the same time, saving me a lot of time. But even if I use the full FOV, I still drizzle. I can easily notice the difference. I tested to see the difference between 2x and 3x cropped images on Crescent (and Horsehead) and there was a minimal but noticeable difference. Some of those filament structures inside the crescent are a few pixels wide and adding a few more helps with how we perceive it-- at least. Which is what we are interested in whatsoever, it doesn't matter if it isn't really recovering something. To make sure I'm not crazy I sent the comparison to couple of friends without telling them which is 2x which is 3x and they were all able to tell.

Sorry that I did not keep this comparison data-- maybe I'll do it again and share the results one day. But I always recommend testing things out yourselves to see what fits best for your setup and workflow.

Arun H

Frequent contributor

Mar 9, 2025

Tony Gondola:
That's what I had assumed. My question wasn't about the limits of what drizzling can do. It was how can you get an image with a final measured average FWHM of under 1.0" with a 2 inch aperture? I must be measuring things incorrectly.

Yes, the FWHM measurement must be made on the linear and not stretched image.

Well Written

Kyle Goodwin

Mar 9, 2025

Also before it’s been improved with BlurX.

Wei-Hao Wang

Mar 9, 2025

First, if one wants to claim drizzle improves the resolution, one really should compare before/after images with the same pixel scale. I do agree that Kay's drizzle result is very impressive from a Cat51. But to really claim that's due to drizzle, a fair and controlled comparison is necessary.

At the same time, measuring FWHM in an image after BXT is scientifically meaningless. But it is indeed a good measure of visual impact. Even if the smaller FWHM doesn't tell the intrinsic resolution of the data, it tells about visual sharpness (along with a potential risk of processing artifact, i.e., sharp features that do not really exist). But again, if you want to claim the small FWHM is caused by drizzle, then you really should measure it before you apply BXT. On the other hand, if one wants to claim the small FWHM is caused by BXT, the he/she should measure the FWHM after BXT on an undrizzled image. If you don't do this controlled comparison, you may attribute the good result to a wrong reason.

All that being said, many people find that BXT works better on images with finer pixel scales (whether it's achieved by drizzle or not). So what Kay saw here may be a combined effect of drizzle+BXT (and excellent processing!).

Well Written Helpful Insightful Engaging

Coolhandjo

Mar 9, 2025

If I dither well, then I will almost always Drizzle 2 x with 0.75 drop Shrink factor unless i'm really oversampled. The result is always better than if I don't.

Wei-Hao Wang

Mar 9, 2025

To OP, if after trying drizzle on a couple of images and not seeing why it makes sense, then not drizzling in the future is a perfect choice.

Drizzle is really only needed when you images are under-sampled. Well sampled data do not need drizzle.

But still, you should dither, always. There is a saying "dither or die" and I think it tells the truth.

Helpful Concise Engaging Supportive

andrea tasselli

Frequent contributor

Mar 9, 2025

Wei-Hao Wang:
But still, you should dither, always. There is a saying "dither or die" and I think it tells the truth.

Nah...

D. Jung

Mar 9, 2025

andrea tasselli:
Wei-Hao Wang:
But still, you should dither, always. There is a saying "dither or die" and I think it tells the truth.

Nah...

This is why we can't have nice things...

Ed Beshore

Mar 9, 2025

Check out my post here

https://www.astrobin.com/8iz85p/

and you'll see a good up close example of how drizzling can help when you are under sampled...

Ed

Andy

Mar 9, 2025

Do you leave your images then it the 2x drizzled size or do you reduce it?
the big file size makes processing considerably slower, at least this is what I feel.

Stephan Linhart

Mar 9, 2025

To chip in my opinion to the conversation using only Color cameras

I always drizzle 2x2 for undersampled data. There will be a drastic improvement in details wen applying a careful deconvolution or apply I may try 3x3 drizzling, however I believe my Computer may not like it. We’ll see.
-With 2500mm focal lenght I am very oversampled in my seeing conditions, I still drizzle 1x1 and get better stars. I believe drizzle helps correcting Bayer artifacts, but correct me if I am wrong. I used this technique before BXT existed ,because I got better results on the stars.

HOWEVER: You will end up with a noisier image, which can be a massive trade off.
That’s why I will try and compare non drizzling + BlurX and 1x1 drizzling with BlurX with my oversampled data to see if it will still be beneficial.
For my Color cam undersampled data I will remain with the 2x2 drizzling, The results are so much better.

@andrea tasselli
Can you please explain your rationale of not dithering a bit more details aside a „nah“.

Has it to do with new cosmetic correction features?
For images where I don’t drizzle it would clearly save time not to dither.

Cheers,

Stephan