3x3 Drizzle Integration

[Atmos]

I know that 9 dithered images is statistically significant enough to do a reasonable job on a 2x2 drizzle BUT what would be required to successfully attempt a 3x3?

I am thinking HORRIBLE sampling like mono imaging with a 14-50mm lens.

[Shiraz]

No idea, but you could try it out. Take a set of well sampled dithered images, run them through software 3x3 binning and then attempt to reconstruct using 3x3 drizzle.

[Atmos]

Sadly the only camera I currently have capable of testing with is my D700 and for whatever reason I cannot get a FWHM lower than 2.8 with it regardless of the scope or lens being used. I get 2.8-3 pixel FWHM at 24, 35, 50, 85 and 680mm. Not sure what it is with my D700 but it doesn't seem to under sample :/

With my QHY22 I am imaging at 1.39"/ pixel so unless I get some seeing like you've had, I cannot really test more than 2x2 drizzle.

[Shiraz]

surely you can? take any good set of dithered subs (maybe use 30? for convenience) that you already have that has reasonably sharp detail - use that as the starting point and stack them to generate a reference image. Then software bin those subs 3x3 to get an undersampled set of subs - and then 3x3 drizzle stack the binned subs to get back to the original scale. compare the drizzle restoration result with the original stack and see how much degradation there has been - and how much detail you can recover. That should tell you how much use 3x3 drizzle is.

your D700 may well have an anti-aliasing filter, which will limit the FWHM?

[Atmos]

That wasn't something I'd thought of so I have gone and done some testing, doesn't quite work as well as I'd hoped :/
I had been trying to figure out why I couldn't get more resolution out of my D700 but I had forgotten about that annoying anti-aliasing filter.

So my discovery so far. I started with a 33x600s stack of Ha on NGC 248, created a group of 2x2 & 3x3 binned images, registered and drizzled them different amounts.

The original FWHM is 2.262 pix.
The 2x2 binned and drizzled 2x2 has a FWHM of 3.175 pix which I found quite surprising as I was expecting it to be closer to 2.4 or 2.5 pix.
The 3x3 binned and then 3x3 drizzled ended up at 4.048 pix.
The 3x3 binned and then 2x2 drizzled and resized to match ended up at 4.174 pix.

I imagine things will work a bit differently in the real world. In my mind under sampling has a different effect to binning when light is hitting the sensor, just in the way flux is recorded and what not. Has been a good test anyway. Will just have to try a real world example at some point into the future.

[RickS]

Your image scale will need to be around the same size as the seeing or greater. To be well sampled you want an image scale around 1/3 of the seeing, so for 3x3 drizzle your subpixels should be at least that scale.

I have some data from DSW in New Mexico from a Rokinon 135mm lens and KAF-8300 with an image scale of about 8.6 arcsec/pixel. It drizzles well at 2x2. I'll give it a go at 3x3 and report back.

[RickS]

Here's a small area of M31. Data captured at DSW with Rokinon 135mm lens and KAF-8300 sensor (image scale 8.62 arcsec/pixel.) First image is 1x1 integration, second is 2x2 drizzle 0.7 drop size and third is 3x3 drizzle 0.7 drop size.

FWHM (pixels) measured with SubframeSelector is 5.172 (1x1), 4.861 (2x2) and 4.73 (3x3) after upsampling the 1x1 and 2x2 to match the 3x3.

The 2x2 is clearly superior to the 1x1. With blink I can see a slight improvement in star size but no other visual improvement going from 2x2 to 3x3.

Cheers,
Rick.

[Atmos]

Thanks for that Rick, how many subs went into that?

I was hoping that the 3x3 drizzle may be able to help even a little but I guess it is not to be :(

[RickS]

Sorry Colin, forgot to mention that. It was 43 x 300s luminance.

Unfortunately, I think the returns are rapidly diminishing above 2x2 but it may be able to squeeze a bit more detail out by playing with drop size and perhaps by tweaking the kernel function (a new feature in DrizzleIntegration.)