Drizzle in PI

[Juan Conejero]

Harry,

I have had a play and I can not get the same improvement in my images ( Like non at all )

I was using sxvf M25 3000 x 2000 colour files and imaged at .8 arc sec per pix

Are your images oversampled? I think this method will work with undersampled data.

[Harry page]

Hi Juan

Yes they are oversampled and was coming to the same conclusion as you ( Honest  )


Harry

[NKV]

I think this method will work with undersampled data.

Juan, you are right. Drizzle improve resolution only at undersampled data.

But, anyway, i tried and i want to show one more result:
I used set of 40pcs 3pix FWHM monochrome images.
1. Registration > Integration > Crop > Deconvolution.
2. Bicubic Spline up-sample 3x > Registration > Integration > Bicubic Spline Down-sample 3x> Crop > Deconvolution.
3. Same as 2. but other combination up-sample 3x / Down-sample methods. Nearest/Bilinear/Integer up sample to any of Nearest/Bilinear/Integer down sample.
Registration, Integration & Deconvolution with identical settings. I use deconvolution in the experiment for self-control.

Conclusion: no undersampled = no improvement in resolution.
But noise... just load attached image in PI, zoom, and compare noise (use Ctrl+PageDowe).
I don't understand what i really lost, but by method 2. look like i lost totality noise.

Maybe Juan can explain what's going on. I just want be sure that this right way.

[NKV]

Second image in attachment.

PS Juan, sorry. I avoid attachment's size restriction.
Sorry for one more post.

[Harry page]

Hi


I agree with my trys there was a slight reduction in background noise , But I think this might be the process working on a higher resolution image ie Noise reduction seemed to work slightley better !

To confirm this I resampled a ordanary image and got the same result  ( this being my oversampled image )

Harry

[Juan Conejero]

My interpretation is as follows. Bicubic spline interpolation has worked here as a convolution, hence as a smoothing filter. This easily explains the low-pass filtering that has removed a good amount of noise: any interpolation algorithm would have produced basically the same result. However the bicubic spline interpolation function has two negative lobes that generate some ringing effect (intentional) to prevent loss of high-frequency detail (that's why this particular interpolation works so well).

It seems that the spline's ringing effect has worked here as an edge protection mechanism, much like edge protection works in ACDNR. The edge protection effect seems to have been very effective to prevent blurring of stars edges, nebular edges, and other high-frequency features. The cause may be in the upsampling of the data. Interesting

[mmirot]

I tried 3X upsampled using nearest neighbor images from The FSQ and PL16803 ( 9u).
Definitely, improved not as striking as the first example.
I wish there was a way to automate this rather than stepwise.
I used image container for the up sampling, then star reg module and integration module.

Also, is there a way to blink images?

Max

[NKV]

is there a way to blink images?

I use Ctrl+PageDowe.

[Niall Saunders]

Hi Max,

Also, is there a way to blink images?

Juan reminded me of an excellent way to do this in (I think) his recent 'Mosaic' video tutorials.

In any case, I will try and describe the method here.

You have two images - effectively the same size. If not, then use DynamicAlignment to make a modified copy of the 'second' image, in such a way that it is 'perfectly' aligned to the first.

On the 'first' image, grab the name tab on the left-hand edge bar of the image, and drop it just below where the name tab was - this will create a full-size preview of the 'first' image

Click on the new preview name-tab - the green border of the 'original' image will disappear (it was there to show you 'where' the newly created preview actually was), and the preview name tab will become a darker grey.

Open an instance of PixelMath and enter a very simple expression, comprising of just the 'name' of the 'second' image. Apply the PixelMath instance to the preview you just created (drag and drop the little 'triangle')

Close PixelMath

The preview now has a 'history' associated with it - the 'first' image, which was then replaced by the 'second' image. You can now click on the little 'Preview Undo/Redo' icon on the Preview toolbar and this will toggle back and forth between the 'before' and 'after' states.

If you had balanced the image intensities before the PixelMath transfer, then you should ONLY see the detail changes.

When you have finished, just delete the preview - the images will not have been altered in any way.

Let us know how you get on.

Hope this helps,

[mmirot]

I was thinking I could use image container and process container to automate this sequence.   

Image set in image container -- resample --- output goes to star registration-- out goes to integration 

I can't figure out to set a reference frame that has not been created yet 

We need some documentation on these features 


Max

[NKV]

excellent way to do this

[Niall Saunders]

Now THAT is a classic example of "a picture painting a thousand words"