Drizzled LMC

[avandonk]

After downloading the latest version of PixInsight with the new drizzle module. I just had to try it.

Data: 30x4 min for each of RGB.

Large image twice camera pixel size. 1.54 seconds of arc per pixel. 30MB

http://d1355990.i49.quadrahosting.co...CF_DRZ_RGB.jpg


Seems to work very well. Note drizzle is only useful if your sensor under samples your optic. Dithering is essential and many sub frames also improves the result.

Here is an animated gif showing the effect. 2MB

http://d1355990.i49.quadrahosting.co...4_06/drzzl.gif

Bert

[SimmoW (SIMON)]

Nice, a very busy area. Now this drizzling, I've seen the updates but am still confused at its use. What do you mean by under samples the optic? When will drizzling be of benefit? Apols, I'm just a beginner. Just doing my first ever 3 min subs right now, unguided, so exciting!

[alpal]

Nice data Bert.
Drizzle is certainly worth doing - excellent demo on that gif file.

I wanted to see what I could do with the brightest central part.
I hope you don't mind me playing with your data?
Do you like the look of this?

cheers
Allan

[mithrandir (Andrew)]

Originally Posted by SimmoW

Nice, a very busy area. Now this drizzling, I've seen the updates but am still confused at its use. What do you mean by under samples the optic? When will drizzling be of benefit? Apols, I'm just a beginner. Just doing my first ever 3 min subs right now, unguided, so exciting!

See http://www.ccd.com/ccd113.html for an explanation of sampling.
Also http://www.cloudynights.com/ubbthrea...0/Main/4215490

[SimmoW (SIMON)]

Thanks for the links Andrew, that clears it up plenty. No need for dithering thankfully (although I bet my DSLR is inefficient and heavily oversampling).

[Phil Hart]

Thanks for the example Bert.

How much better do you think the drizzled result is than if you had just resampled the un-drizzled data to the same resolution? It looks like you'd get a similar result just through interpolation. Not sure how much more is being achieved by the drizzling?

Phil

[alpal]

Originally Posted by philiphart

Thanks for the example Bert.

How much better do you think the drizzled result is than if you had just resampled the un-drizzled data to the same resolution? It looks like you'd get a similar result just through interpolation. Not sure how much more is being achieved by the drizzling?

Phil

This example from DSS explains it well:

http://deepskystacker.free.fr/english/technical.htm

What and when you need to use the drizzle option
Basically you need only to have a lot of images which must not be perfectly aligned (a few pixels drift is enough).
It makes no sense to drizzle if you have only a few light frames.

Drizzling is particularly good at enhancing the resolution so it is very useful when shooting small objects with a short focal length.

On the right is a example of an almost unprocessed M57 (shot using a 10", F/D 4.72 OTA with a Canon DSLR).

Usually M57 is very small, but by using a 2x drizzle and around 100 light frames the resulting image is twice bigger and the resolution is much better.

Mouse over to see the

Hold the mouse over the 2 links to see the difference.

[avandonk]

Originally Posted by philiphart

Thanks for the example Bert.

How much better do you think the drizzled result is than if you had just resampled the un-drizzled data to the same resolution? It looks like you'd get a similar result just through interpolation. Not sure how much more is being achieved by the drizzling?

Phil

The stars seem to be the about the same with upsizing and stacking. The problem with upsizing and stacking is that a lot of interpolation of signal is invoked before distribution to the finer mesh..

With drizzle there is NO a priori interpolation. The process is to distribute real samples of intensity from a coarse mesh onto a finer mesh. This is only possible because of the fact that the coarse mesh was dithered. What is happening is that the extra spatial information due to the dithering is being traded for greater resolution for an increase in noise.

In our Universe there is no free lunch!


I am sure that the faint nebular detail is far better with drizzle. This NB Hubble Palette of the LMC shows far better faint detail with drizzle than with upsizing and stacking. I have just reprocessed the data.

Large image 30MB

http://d1355990.i49.quadrahosting.co...LMC_DRZ_HP.jpg

The other advantage is that at the calibration stage I am converting the 16 bit data subs (32MB) into 32 bit FP images (65MB). This would be computationally difficult with upsizing as a 2x 32 bit FP image is 262MB in size.

At the alignment stage PI uses the sensor size 32b FP images to produce aligned images and a corresponding .drz file with the 'spatial' information.

Integration is then done with these pairs of fit and drz files which produces a much larger updated drz file. It is at this stage that rejection can occur with the original data not interpolated data.

These now much larger drz files are used as data by a module called DrizzleIntegration to produce the final drizzled file which is 262 MB in size ie x2 32b FP.

Alignment and integration is far faster with drizzle than with upsized files. I was only upsizing by a factor of 1.5. My computer systems resources would choke with files upsized by a factor of 2 even at 16 bit. My system is a four core I7 with 12GB of RAM.


My brain cells are getting far more exercise than my body lately.

I forgot to say the blokes at PixInsight are very smart!

Bert

[avandonk]

Here is just the NII data drizzled. Both 18MB

http://d1355990.i49.quadrahosting.co...MC_NII_DRZ.jpg

Inverted

http://d1355990.i49.quadrahosting.co...II_DRZ_INV.jpg


The faint stuff is ...


Bert

[cosmophoton (Luiz)]

Nice examples, Bert, thanks for sharing!
Luiz

[SimmoW (SIMON)]

Hey Bert, what scope n mount do you use? A buddy just asked me after I sent him this great image.