PixInsight

[multiweb (Marc)]

Can you use Bayer drizzle outside the BatchPreprocessing script?

[lazjen (Chris)]

I have no real idea, but would this forum post help: http://pixinsight.com/forum/index.php?topic=8232.0 ?

[multiweb (Marc)]

Originally Posted by lazjen

I have no real idea, but would this forum post help: http://pixinsight.com/forum/index.php?topic=8232.0 ?

Thanks Chris. Found this: http://pixinsight.com/forum/index.php?topic=8319 Installed the new module and going through the steps now.

[multiweb (Marc)]

Well, it's a little tedious but the result is worth the time. Left is drizzle integration (4:1) and right is standard integration (8:1). Apparently the next release of PI will have bayer drizzle out of the pre processing script so you can do it on already calibrated subs. Until then we have to use that module and do it in more steps.

[peter_4059 (Peter)]

That's a pretty amazing improvement in the look of the stars.

[multiweb (Marc)]

Originally Posted by peter_4059

That's a pretty amazing improvement in the look of the stars.

Yes it does. Pretty amazing the resolution you get back. Effectively double it.

[Garbz (Chris)]

For completeness sake would you be able to post the comparison against the non-drizzle image resized to 200% ?

A fair comparison would be using Drizzle to restore undersampled data vs using interpolation to restore it. No doubt the drizzle will win but I'm wondering how much of a difference it would make.

[Atmos (Colin)]

Originally Posted by multiweb

Yes it does. Pretty amazing the resolution you get back. Effectively double it.

I read a paper on this a little while ago, from memory (so let's say a ball park figure!) it encroaches nearly the 190% mark. So although the image is effectively doubled, under perfect circumstances you can only get close to double but not reach it.

[multiweb (Marc)]

Originally Posted by Garbz

For completeness sake would you be able to post the comparison against the non-drizzle image resized to 200% ?

A fair comparison would be using Drizzle to restore undersampled data vs using interpolation to restore it. No doubt the drizzle will win but I'm wondering how much of a difference it would make.

True. At 200% it looks similar.

Originally Posted by Atmos

I read a paper on this a little while ago, from memory (so let's say a ball park figure!) it encroaches nearly the 190% mark. So although the image is effectively doubled, under perfect circumstances you can only get close to double but not reach it.

Fair enough.

[peter_4059 (Peter)]

Which one is drizzled?

[multiweb (Marc)]

Originally Posted by peter_4059

Which one is drizzled?

Right one is drizzled. You can see the left is pixelated slightly.

[Camelopardalis (Dunk)]

Whoa, that's quite a difference

[Garbz (Chris)]

Wait does drizzle increase the resolution by 200% each way or 400%? The left one looks lower resolution than the right.....

[Atmos (Colin)]

Originally Posted by Garbz

Wait does drizzle increase the resolution by 200% each way or 400%? The left one looks lower resolution than the right.....

Drizzle will turn a 512x512 into a 1024x1024.

[multiweb (Marc)]

Originally Posted by Garbz

Wait does drizzle increase the resolution by 200% each way or 400%? The left one looks lower resolution than the right.....

Set to 2 in PI so 200%. Never tried other settings. Yes, left is zoomed past the pixel size so 2:1, right is drizzled displayed at 1:1 as it's double the size.

[Garbz (Chris)]

So what if you resize the original to have the same pixel size as the drizzle image? That's what I was originally getting at. Same final resolution, but drizzle vs interpolation.

[multiweb (Marc)]

Originally Posted by Garbz

So what if you resize the original to have the same pixel size as the drizzle image? That's what I was originally getting at. Same final resolution, but drizzle vs interpolation.

It's really hard to tell which one is better. The details are quite similar. The stellar profiles and star edges are better in the drizzled version. I can also see that the noise level in the background seems to be slightly lower when drizzled.

TBH I have had mixed results. Bayer drizzle in PI didn't work consistently. I did another debayering in CCD Stack then did a two step integration in PI to generate *.drz files on individual channels. It seems to work better. Will have to try MAXIM DL which also has drizzle in the latest version and see what happens there. Will post results when I have all the stacks to compare. Apparently Nebulosity does it as well so does DSS I hear.