Drizzle all the time

[Larry Phillips]

I have been reading comments here and on other sites that the considered preference is to use drizzle in WBPP for all images regardless of the plate scale. Is this true for mono and OSC? m If it is true, what is the advantage for short focal length images.

Larry

 

[cloudbait]

I have been reading comments here and on other sites that the considered preference is to use drizzle in WBPP for all images regardless of the plate scale. Is this true for mono and OSC? m If it is true, what is the advantage for short focal length images.

Larry

I virtually never drizzle my mono images, which are natively oversampled. I've experimented, and there's absolutely no improvement in image quality (just larger files that require more storage and more processing time). The general consensus is that 1X drizzle is useful for OSC images.

 

[fredvanner]

x2 (or higher) drizzle only if undersampled (test: FWHM < 2 pixels) and plenty of subs.
x1 drizzle only if OSC and plenty of subs.

 

[ngc1535]

x2 (or higher) drizzle only if undersampled (test: FWHM < 2 pixels) and plenty of subs.
x1 drizzle only if OSC and plenty of subs.

This is not what Vicent's video says. He shows using drizzle all the time at 1x1 even on monochromatic data with small numbers of subframes at 1x1. In the video there are just 20 frames.

@Juan Conejero Could you please justify this statement and example shown? I understand the need to do CFA drizzle at 1x1 for OSC.
Obviously other experts on this list (myself and the above) are not aware of this either- thus an explanation is warranted.

Thanks,
Adam

 

[fredvanner]

I understand the need to do CFA drizzle at 1x1 for OSC.

It is not a need, it is a minor benefit if you have enough subs. If you don't it may produce artefacts.
There is absolutely no benefit that I am aware of from using x1 drizzle with mono subs. If I am wrong about this, I would like to see an explanation of why.

 

[ngc1535]

It is not a need, it is a minor benefit if you have enough subs. If you don't it may produce artefacts.
There is absolutely no benefit that I am aware of from using x1 drizzle with mono subs. If I am wrong about this, I would like to see an explanation of why.

Please review the video. Let us make certain this just isn't an error in my ability to understand what is implied in the video.
It appears to imply we are missing something important. The video is at the timestamp it is stated.
-adam

 

[johnpane]

There is absolutely no benefit that I am aware of from using x1 drizzle with mono subs. If I am wrong about this, I would like to see an explanation of why.

I would think that if the dropsize is less than 1 there could be spatial resolution benefits even with monochrome images.

 

[ngc1535]

I would think that if the dropsize is less than 1 there could be spatial resolution benefits even with monochrome images.

*Always* ...regardless of number of images?
Guessing just means we need an explanation. (John.. you are a smart guy!)
-adam

 

[johnpane]

*Always* ...regardless of number of images?
Guessing just means we need an explanation. (John.. you are a smart guy!)
-adam

"not never" does not mean "always".

The drizzle algorithm was invented for the Hubble Deep Field, which was captured using monochrome sensors (Fruchter and Hook, 2002).

 

[cloudbait]

"not never" does not mean "always".

The drizzle algorithm was invented for the Hubble Deep Field, which was captured using monochrome sensors (Fruchter and Hook2, 2002).

It was developed because the HST camera undersampled the field. Nobody is suggesting that drizzle isn't useful with a mono camera when it is undersampling. In which case, you use 2X or higher (not 1X).

In the case of OSC cameras, the algorithm is serendipitously doing something that it was never developed to do when used at 1X. But only OSC cameras.

 

[johnpane]

It was developed because the HST camera undersampled the field. Nobody is suggesting that drizzle isn't useful with a mono camera when it is undersampling. In which case, you use 2X or higher (not 1X).

You are equating the improvement on undersampled images to the scale factor. Consider that a 1x drizzle with a dropshrink<1 can also counteract undersampling.

 

[ngc1535]

You are equating the improvement on undersampled images to the scale factor. Consider that a 1x drizzle with a dropshrink<1 can also counteract undersampling.

Just to keep this on track. The issue (in my mind) is that the video clearly says ALWAYS and does so in the example of monochrome data of unspecified sampling and small number of frames.

-adam

 

[cloudbait]

You are equating the improvement on undersampled images to the scale factor. Consider that a 1x drizzle with a dropshrink<1 can also counteract undersampling.

The improvement (as the algorithm was designed) is in scale factor!

I'd suggest doing what I did and experiment. With my moderately oversampled data, I can discern no improvement either visually or looking at image statistics.

 

[vicent_peris]

Hi,

Drizzle (including x1) will avoid the artifacts of the interpolation algorithms we have in PixInsight in both OSC and monochrome images. These artifacts include the tendency of the stars to become diamond-shaped and ringing around stars with very small PSF (though in this case it is preferable to use drizzle x2).

Drizzle will always give a resulting master light with a healthier noise distribution: interpolation algorithms tend to displace the noise towards larger scales and blur it in areas with aliasing. In plain sight, without any noise reduction applied, the visual appearance of noise with drizzle will be always more comfortable.

Lastly, drizzle usually works very well from 15-20 subframes and, in most cases, it is preferable to the interpolation artifacts. This is not a fixed or exact number, though. One should always perform an inspection of the resulting master to be sure everything went well.


Best regards,
Vicent.

 

[ngc1535]

Hi,

Drizzle (including x1) will avoid the artifacts of the interpolation algorithms we have in PixInsight in both OSC and monochrome images. These artifacts include the tendency of the stars to become diamond-shaped and ringing around stars with very small PSF (though in this case it is preferable to use drizzle x2).

Drizzle will always give a resulting master light with a healthier noise distribution: interpolation algorithms tend to displace the noise towards larger scales and blur it in areas with aliasing. In plain sight, without any noise reduction applied, the visual appearance of noise with drizzle will be always more comfortable.

Lastly, drizzle usually works very well from 15-20 subframes and, in most cases, it is preferable to the interpolation artifacts. This is not a fixed or exact number, though. One should always perform an inspection of the resulting master to be sure everything went well.


Best regards,
Vicent.

Great. Lets nail this down.From the above I understand Drizzle 1x1 is nearly always best (a slight digression from the video). Furthermore:

1. Drizzling 1x1 is recommended with as few as 15 frames and of any sampling. This is certainly new information and the first time I can recall it being advised. If you have documentation that shows this advice previously, please share it.
2. No limits. are indicated. Suggested is to "inspect" the resulting master. But what should we look for? What are the boundary conditions? Should we expect or look for holes? Is there a difference between drizzled 1x1 oversampled data and non-drizzled with Lanczos interpolation in terms of SNR. Mentioned is that the noise will have a different "distribution"- but it is not clear how this relates to SNR photometrically. Is dithered data a prerequisite?

Please do outline not only the positives- but also the things to look out for when things "do not go well". When thousands of users take this advice the boundary conditions matter and are significant.

-adam

 

[Juan Conejero]

Hi Adam, to answer quickly to your questions:

1. We recommend using drizzle on a regular basis when at least 15 well-dithered frames are available. x1 if well sampled, >= x2 if undersampled.

2. Just look for possible dry pixels (holes). Drizzle weight maps are helpful for this task. The result should be fine if the frames are well dithered.

 

[rberrios1]

Does the intention of using blur XT have any say, or sway in this decision?
Ross does say something about this, although vaguely, in his YouTube video in the Astro channels.
If I am sampling at .71. With scene between 1 and 1.5. Would I benefit from drizzling 1 X before using blur XT?

 

[ngc1535]

Hi Adam, to answer quickly to your questions:

1. We recommend using drizzle on a regular basis when at least 15 well-dithered frames are available. x1 if well sampled, >= x2 if undersampled.

2. Just look for possible dry pixels (holes). Drizzle weight maps are helpful for this task. The result should be fine if the frames are well dithered.

Thank you Juan.
I just need to be as informed as possible...before I go march off and advise many people...

-adam

 

[Scott Badger]

FWIW, I image at 0.33” and the seeing limits me to 2.25” fwhm at very best, but most often a lot worse, so oversampled and then some by anybody’s standards. I have a mono camera (268M) and dither, and always at least 30 subs for the integration. I use drizzle at 1x and though I’m not qualified to speak to the theory, I can report that drizzling at 1x decreases the noise and increases the SNR in my masters according to the image analysis tools in PI. I don’t believe I’m seeing ‘dry’ pixels due to drizzling, but if that were happening, what impact would they have on the metrics?

Cheers,
Scott