Drizzling, oh my!

[Droogie 2001]

I have been using PixInsight for about 2 months now and have been going through some old data.

Having become a bit more familiar the program I decided to have a go at Drizzling my data.
Following the steps listed in the ever-helpful Light Vortex website I was astounded by the results.

The image below is a comparison of the drizzled image (x2) on the left and the exact same data on the right with no drizzling.
I realised my set-up was going to be under sampling my images from the start with a 3.17 arc seconds per pixel rating but I had not pay too much attention to this.
I was also wondering why some of my stars in post processing were ‘bleeding’ colour when the RGB channels were combined, I now believe this was due to the blocky stars. I haven’t really got much further with my new drizzled data yet so I do have a couple of questions about the drizzled images before ploughing on.

I would imagine there is a payoff for drizzling images and not just the fact it takes longer to render and uses up more disk space.
Firstly the actual size of the drizzle images are larger (twice I guess, I selected default of x2?) Is this going to lower the resolution I had read that it doubles the resolution as well?

Can I leave them drizzled for the rest of the processing? Would there be any reason to reduce them back to their original size afterwards?

I also understand drizzled images benefit from good dithering. This is something I will be including in my workflow over the coming months when the Nebulas are back in town.

Needless to say, my appreciation for PixInsight keeps growing. I would imagine if you could buy a piece of optical hardware that could drizzle your star’s automatically people would pay a small fortune for it.
The fact that this ability is bundled within PixInsight makes its value ever greater.

Thanks.

[frugal]

Normally when you stack images it works out how far out each image is compared to a reference image (say -2.34x, +1.93y, 0.02 degrees rotated). The software then uses this transformation to remap the image onto a new image that aligns with the reference image. In the process of doing this, it will not be able to fit the source pixels exactly into the destination pixels, so it does the best it can, this shuffles the data around a fraction before stacking. Drizzle integration keeps the original image, along with the transformation matrix and at integration time it calculates what percentage of the source pixel will cover the target pixel and uses that value rather than an all/nothing approach. The x2 drizle doubles the size of the target image so that your source pixel now covers more target pixels (somewhere between 4 and 9 for a x2 drizzle). The drop shrink value scales down the size of the individual pixels from the source image so that they are only 90% as big (0.9 dropshrink) so that the data does not bleed over into more destination pixels than is desired.

For DSLRs and other OSC cameras it has the added advantage that it can be applied to the individual R, G and B frames to try to restore some of the data that is not present in each of the layers (R is only 1 pixel in 4, so 75% of the R layer is completely black). In this case I tend to use a scale of x1 and a drop shrink of 1 as I am only using 25% of the data from each frame.

The nice thing about drizzle is that with enough frames you can work on targets that do not take up much of the field of view and still have them at a reasonable size once you have cropped them

With regards to the questions: 

 - Resolution - If you drizzle a bunch of 1000x1000 images at x2 drizzle, then you will get a 2000x2000 pixel image. In this case your arc seconds per pixel has now changed from 3.17 to 1.56

 - Rescaling the image - Drizzling is the name for the integration process, it is just a greyscale or RGB image at the end of the process, the rest of the image processing chain can continue as normal (just remember that you are no longer as oversampled as you were so things like deconvolution may no longer be needed / work). There is no need to resize the image back down to the original size unless you are finding it too slow to process. If the example 2000x2000 is too large to work with, then just use a drizzle x1 in the first place.

- Dithering - The reason that dithering helps drizzling is that the image falls on slightly different pixel boundaries on each frame. I.e. the centre of a star might be in the middle of a pixel in one image, then off to the right on the next image, and up at the top left of the third image. This will mean that the centre pixel and all of the surrounding ones have subtly different intensities, drizzling helps use this information to get more out of the information you already have.

I do not know with LRGB if it is worth drizzling all 4, or just the L

[Droogie 2001]

Thanks Frugal for the comprehensive response.

So I am not losing any resolution by Drizzling? It almost seems to good to be true...
I had not considered the benefits of targeting smaller objects. Being a wide field refractor limits me in some areas, this could mean some galaxies / smaller nebulas are potentially viable targets as I will not suffer so much lose if I crop them?

If I reduce the drizzling to say x1 then I guess that will mean the stars will not be as round as say x2 i.e. Less Blocky?

Thanks.

[frugal]

Drizzling only works when you have more information in your source files than you need to integrate the image. So you are not gaining data, you are making more use of the data that you do have. When you do the Drizzle Integration, the drizzle amounts get shown in a new image. If it is a nice solid green, then you have enough data to drizzle at that resolution. If you get a speckled or checkerboard effect when you look at it a 1:1 then you do not have enough data to fully drizzle your image. Either gather more data, or go to x1 rather than x2 drizzle.

At x1 drizzle you will have the same number of pixels that you started with. Obviously there is only so small you can go with a star before it reaches a 2x2 block of pixels.

[Droogie 2001]

Thanks again.

Is this the Drizzle Weights image that is outputted at the same time as the new drizzled image is generated?

Just ran a quick test on x12 Red subs and it produces a white image. Don't have time at the moment to re-run on the Luminance of which there are 47 frames. I think I recall a green image when I ran this?

The Red outputted image still looks much better than the none drizzled image? Does not seem to be any difference to stars at either x1 or x2 both look much smoother than the non drizzled. So presumably it would be still okay to use these drizzled images even if there are are not as many sub frames as the Luminance?

[frugal]

4 minutes ago, Droogie 2001 said:

Is this the Drizzle Weights image that is outputted at the same time as the new drizzled image is generated?

Yes, this is the Drizzle Weights image. 

[ollypenrice]

Regarding the two images you posted at the top of the thread, they are certainly not identical. The crop is significantly different, the right hand image being shifted upwards and to the right of the one on the left.

I'm interested in this issue, though, since in the last few years I've neen undersampling, so drizzle ought to help me boost resolution.

Olly

[Droogie 2001]

The sub data is identical. I ran one with drizzle and one without. The images I pasted were manually moved by hand / mouse to show the same stars in the same area. It wasn't done with any accuracy it was just to demonstrate the stars came out rounder and not blocky.

The drizzled image size does come out larger (x2 larger) so not sure if this is what you are referring to. I believe the stars look better in the drizzled image in my opinion.

 

[wimvb]

2 hours ago, ollypenrice said:

Regarding the two images you posted at the top of the thread, they are certainly not identical. The crop is significantly different, the right hand image being shifted upwards and to the right of the one on the left.

I'm interested in this issue, though, since in the last few years I've neen undersampling, so drizzle ought to help me boost resolution.

Olly

It certainly ought to. I have found that drizzled images are easier when doing deconvolution and star reduction as well. But it may be better to crop the image before doing extensive processing. The increased size (4 x MBytes for a 2 x drizzle)  makes for longer processing times (at least for PixInsight on a laptop).

While drizzled images need some form of dithering, it is the fraction of a pixel offset that is needed. If dithering were an exact number of pixels, drizzle wouldn't work. If tracking/guiding isn't 100 % spot on, there may be enough offset between subs even when no active dither is used.

[ollypenrice]

8 hours ago, wimvb said:

It certainly ought to. I have found that drizzled images are easier when doing deconvolution and star reduction as well. But it may be better to crop the image before doing extensive processing. The increased size (4 x MBytes for a 2 x drizzle)  makes for longer processing times (at least for PixInsight on a laptop).

While drizzled images need some form of dithering, it is the fraction of a pixel offset that is needed. If dithering were an exact number of pixels, drizzle wouldn't work. If tracking/guiding isn't 100 % spot on, there may be enough offset between subs even when no active dither is used.

Also few, if any, people have a PA delivering sub pixel accuracy over a run.

Olly