Drizzle Combine

[Nigel Ball]

Hi

Has anybody written an add on to perform Drizzle Combine in PI?

Nigel

[Niall Saunders]

Nigel,

Maybe my brain is fried from having just driven a 750-mile round trip for a two-hour business meeting yesterday ( ), but I cannot see how 'Drizzle Combine' is any different from 'normal combine'.

Surely, the point of 'Drizzle' actually takes place at the 'capture' stage, not at the point of 'combination'?

After all, no matter how you set about acquiring the data, at some point you have to pick a reference image (real, or synthesised) and use that to align all the other images in your data set.  You can't really alter that fact. After all, any image that is not 'perfectly aligned' would surely then simply be contributing TO the 'noise element' of the resultant image. In other words (again, no matter 'how' the original images were acquired) you MUST strive for 'perfection' in alignement of all images that will subsequently be stacked.

So, StarAlignment (or equivalent) solves that requirement - perfection of alignement. And then ImageIntegration sorts out (and very well, I might add) the need to 'stack' all the images thereafter.

In my mind, 'Drizzle' - a process applied during the 'capture' stage - serves to either :-

1.) Increase the overall pixel count of a final image - by effectively creating a 'mini-mosaic' of superimposed images, created by allowing the scope to 'wander around' some central reference point, acquiring as many images as possible to allow a decent mosaic to be recosntructed
2.) Decrease the 'noise' impact of 'chip defects' (including dust effects) by ensuring that all images place the key alignment points (such as stars, etc.) at slightly different locations on the CCD, for every image captured - which, as a result of subsequent 'averaging' of the then aligned images, causes this 'noise' to be eliminated because it now has a 'random' spatial relationship to the actual photonic data captured (random noise in 'n' images can, theoretically, be reduced by a factor of '¹/_n-squared')
3.) Increase the 'perceived pixel resolution' of a given CCD sensor due to the effects of sub-pixel-sampling required in the re-alignment process, when the images have been acquired with a random dither applied between images - for a far clearer explanation, have a look at the two articles by Stephen Hamilton at http://www.hamiltonastronomy.com/ASTRO_ARTICLES/DSI_Drizzle_Part_1.pdf and http://www.hamiltonastronomy.com/ASTRO_ARTICLES/DSI_Drizzle_Part_2.pdf (although these are related to the Meade DSI and Meade's Envisage software, they explain things far more clearly that I could ever hope to!!)

So, have I missed the point of your question? I don't really see the need for a "Drizzle Combine" add-on. Can someone else comment?

[mmirot]

Perhaps Juan can speak to this.
Drizzle requires a good dither and a specific resampling process.

 It might be 2x and nearist neighbor resample.
It take a lot of extra time. It can be worth it if you have enough images that are very undersampled. Perhaps 15 -20 or more
This would be good one for simple script.

Max

[Carlos Milovic]

I see the benefit from this "method" for bayered images, or planetary targets... I think that specific tools for those kind of images are required, rather than a general dither process.

[Nocturnal]

Drizzling is described in various Googlable locations but this is how DSS does it: http://deepskystacker.free.fr/english/technical.htm#Drizzle

Personally I've not seen great benefit from doing it. I tried it again with a recent M51.

[Niall Saunders]

Hi Sander,

That explanation (from the DSS site) is very much along the lines of Stephen Hamilton's description (as I remember it).

In which case, the 'up-sample' required prior to alignment is actually very easy to achieve in PI, by simply using the Resample process and an ImageContainer. Then, I assume, you simply align and stack the images as per normal, finally downsampling (as needed) at whichever point you choose.

Yes, it could be argued that 'a script' would help - but the actual work needed is so trivial that 'a script' is just not necessary - use an ImageContainer, it will do all the work for you, AND you can save the container 'setup' as a ProcessIcon (either individually, or as part of an overall 'group' of icons, along with the Resample process icon).

Am I still missing anything?

[jmtanous]

Hi,

One possible solution is:

-- Calibrate your lights.
-- resize your lights
-- Align the new resized lights and continue your normal workflow

The disadvantage of this method over drizzle is that you will use in your stacking process interpolated values. While the real drizzle method just used real values. The problem with drizzle is that you need many many dithered lights to avoid 'holes' in your final image.

For undersampled images, drizzle can make a difference.

Cheers,

Jose

[mmirot]

I tried with my FSQ 106 images. I think am at 5 arc sec/ pix. It does make a noticible improvement in resolution on images 25 more subs. You can imagine this takes some time to do.
I am start with 32 MB file increase the file size with a 2x resample etc.
I was hoping that some day this would be part of processing sequence like it is implemented in DSS and MaxIM DL. 

Max