what do you drizzle at

[iwols]

just found drizzle just wondering at what scale you drizzle i have the option of 0.1 to 3 cheers

[vlaiv]

I think that drizzle should not be used by majority of amateur setups.

In order for drizzle to work properly - one needs quite a bit of under sampling. I think that one also needs exceptional pointing precision. Drizzle algorithm was developed for Hubble if I'm not mistaken. Hubble has what it takes to make use of it - large under sampling - no atmosphere to blur the image and only blur comes from Airy disk. It is also very precise in pointing.

Most amateur setups don't get above requirements for it. You need relatively large pixels compared to focal length and you need large aperture - this means fast system and good guiding and steady skies. Most amateur systems that are fast enough - don't have diffraction limited optics - that will cause additional problems. In addition every subframe should be sub pixel dithered - that is not problematic I think, but dithers are quite random and drizzle works the best on very precise sub pixel dithers.

There is also matter of registering images - for dither to work the best one should avoid interpolation algorithms when registering (this is where precision dithers come in place - you don't need to shift images in algorithm if your dithering is precise enough).

This is what I think anyway, I'm not sure if I have enough solid knowledge about it to make above a claim, so best leave it as an opinion.

[michael8554]

6 hours ago, iwols said:

Just found drizzle just wondering at what scale you drizzle i have the option of 0.1 to 3 cheers

Do you mean Dither ?

What software are you using ?   I thought Drizzle settings were in whole numbers eg  2X,  3X etc.

Michael

[DaveS]

@vlaiv, when you specify undersampling, is that WRT the native resolution of the telescope, or the seeing-limited resolution?

I'm thinking of a non-hypothetical 12" ODK at 2040mm and a 16200 camera with 6 micron pixels, or should I look for even bigger pixels? This would drive me into the 36mm square chips and even more serious money.

[vlaiv]

10 minutes ago, DaveS said:

@vlaiv, when you specify undersampling, is that WRT the native resolution of the telescope, or the seeing-limited resolution?

I'm thinking of a non-hypothetical 12" ODK at 2040mm and a 16200 camera with 6 micron pixels, or should I look for even bigger pixels? This would drive me into the 36mm square chips and even more serious money.

Undersampling can relate to both telescope resolution and compound resolution for ground based telescopes (seeing + scope resolution + guiding/tracking errors).

It stands for having larger pixels than those needed to properly record resolving power of the system. Imagine you have two close stars and under given conditions you can resolve them - you "see" separation. Now image pixel big enough to cover both stars at the same time. This pixel will not be able to record separation between the stars. This is what under sampling is.

Drizzle is method used in above scenario to create stack of images still showing to distinct stars and separation between them. How does it do it? It exploits the fact that stars are never pin points of light due to resolving power of system - they are always having shapes similar to Gaussian curve - or Moffat curve. In case of telescope in space with no atmosphere - it will be Airy curve.

This means that, depending on pixel placement you will get star light on adjacent pixels. First pixel will get for example 90% of light if it covers most of the star, and one next to it will get 10% of the light. If you move scope just a tiny bit - this ratio of star light will change. If star image falls exactly on edge of two pixels - both will record 50% of the light. This actually works in 2d - as you can move scope in horizontal as well as vertical direction and it's group of pixels that receive portion of light.

Algorithm works by taking images each with different offset, and with "clever" method it extracts what sort of image would give different ratios of pixel values based on different shifts. This is why sub pixel offsets between subs are needed to properly do it.

In case you mentioned - you don't have to worry about under sampling, it is opposite that you need to address - it's the over sampling. This is case when you use too many pixels to record star profile (or image in general). In itself that is not a bad thing - you don't loose detail of the image, but it has particular drawback that is of great concern for every amateur imager - light spreads over more pixels and each pixel gets smaller fraction of light - meaning each pixel gets smaller signal, and smaller signal means poorer SNR. This is why it's not good to over sample. Under sampling looses detail, over sampling looses SNR.

Goldilocks applies here - best sampling is proper sampling

 

 

[DaveS]

Thanks Vlaiv

Using the CCD calculator in Astronomy Tools I get 0.61"pp binned 1x1, hence a very doable 1.21"pp binned 2x2 with (Effectively) 12 micron pixels. And without breaking the bank!

[vlaiv]

1 minute ago, DaveS said:

Thanks Vlaiv

Using the CCD calculator in Astronomy Tools I get 0.61"pp binned 1x1, hence a very doable 1.21"pp binned 2x2 with (Effectively) 12 micron pixels. And without breaking the bank!

That sounds like very good option - 1.21"/px

[michael8554]

3 hours ago, vlaiv said:

Algorithm works by taking images each with different offset, and with "clever" method it extracts what sort of image would give different ratios of pixel values based on different shifts. This is why sub pixel offsets between subs are needed to properly do it.

So drizzle doesn't have to be in whole numbers of pixels ?

Michael

[iwols]

not sure about the calculation im using a 580mm with a 414ex ccd with the option of using a 0.8 reducer if anyone can work it out cheers

[vlaiv]

1 hour ago, michael8554 said:

So drizzle doesn't have to be in whole numbers of pixels ?

Michael

Actually, I don't fully understand how drizzle works in amateur setups. I did read an article once and got basic understanding of what is it about.

Method is as follows: Each pixel is considered to be a little square and then it is "shrunk" in size. If we consider image to be grid of squares, then value of each pixel ends up in smaller square in each grid cell. Area around that square is left "empty" - or not having a value.  Something like this diagram:

That leaves space around each pixel as empty. Image depicts 3x3 drizzle. Other subs are done the same, but subs are dithered by subpixel shift of 1/3. This can be done by exact pointing of HST. Once that is done, next subframe will be aligned so that their shrunken pixels fall into empty space of this one. In order to fill the image, you will need 9 subs, each offset with respect to other. This "spent" 9 subs but did not do any stacking, no SNR improvement. In order to do that, you need further frames - each one will be offset and match to particular pixel position - these then get stacked.

This is original drizzle algorithm. It cannot work in amateur setups simply because there is no theory to point and guide scope with such sub pixel accuracy. Here is excerpt from actual Drizzle Handbook found here:

http://www.stsci.edu/hst/HST_overview/documents/multidrizzle/multidrizzle.pdf

" to improve sampling of the PSF, together with increased spatial resolution, the images need to be shifted by sub-pixel amounts. Gener-ally, subsampling by 1/2-pixel offsets provides the most dramatic improve-ment over non-dithered images. In some cases, observers wish to further explore the limits of the instrument and spacecraft pointing accuracy by con-sidering sub-pixel shifts of 1/3 of a pixel or less"

Answer to your question would then be - drizzle works by integer steps 1/2, 1/3, 1/4 of original resolution. So it's not in whole number of pixels but rather whole number of divisions of a pixel.

[vlaiv]

1 hour ago, iwols said:

not sure about the calculation im using a 580mm with a 414ex ccd with the option of using a 0.8 reducer if anyone can work it out cheers

With scope at native focal length, you are sampling at 2.28"/px - that is perfectly fine resolution for small 3-4" scope. You might be under sampling in some conditions, but in average seeing you will be extremely close to optimum sampling rate if using 3" class instrument.

With x0.8 reducer, you are working at 2.87"/px. This will be slightly under sampling for all but the worst seeing.

Both of said resolutions are perfectly fine for wide to medium field imaging, really. You are slightly under sampling in some circumstances, but that is to be expected from wider setups. I certainly would not categorize 2-3"/px as something needing drizzle (even if possible) to recover detail, so you don't ever need to think or worry about it.

[iwols]

4 hours ago, vlaiv said:

With scope at native focal length, you are sampling at 2.28"/px - that is perfectly fine resolution for small 3-4" scope. You might be under sampling in some conditions, but in average seeing you will be extremely close to optimum sampling rate if using 3" class instrument.

With x0.8 reducer, you are working at 2.87"/px. This will be slightly under sampling for all but the worst seeing.

Both of said resolutions are perfectly fine for wide to medium field imaging, really. You are slightly under sampling in some circumstances, but that is to be expected from wider setups. I certainly would not categorize 2-3"/px as something needing drizzle (even if possible) to recover detail, so you don't ever need to think or worry about it.

thanks thats good news cheers