How to stack images of Saturn?

[TrickyDicky]

Warning - Newbie Here !!!
Is it possible to stack images of Saturn? There are no stars on the images.

[Geoff]

At the risk of being shot down in flames, let me suggest Registax. It's freeware and is really good for planetary stacking and processing. Another alternative is to use the fourier transform stacking method in CCDStack

[Juan Conejero]

In PixInsight you have the FFTRegistration script that I wrote some years ago. From the main menu, select Script > Utilities > FFTRegistration.

For planets, don't enable rotation and scaling corrections because most likely they won't work. You have to select one of the images as the registration reference. Ideally, you should choose one of the best frames. If you want to keep the registered images, check the 'Write registered images and select an output directory. If you leave the 'Integrate' option enabled, you'll get the mean of all registered images as a new image window.

Good luck!

[TrickyDicky]

I selected 15 files, one of them as the REF.

The process stopped on the registration of the first image:
143: Error : Out of memory

The images have 6000x4000. Are they to big?

[TrickyDicky]

Error [000]: C:/Program Files/.../FFTRegistratio.js
line 676: Error: out of memory

[Carlos Milovic]

Instead of the script, use the process I wrote
Here are my development modules: http://pixinsight.com/forum/index.php?topic=2275.0
The one you need is this one: http://pteam.pixinsight.com/pcldev/carlos/x64/CMGeneral-pxm.dll
(assuming you are using a 64bits windows)

You may use that process with an image container to batch process the images.

Please note that Fourier based registration is not very robust, specially with heavy noise. In the process you have a wavelets filtering option that may aid to better align the frames, and achieve higher accuracy. Also, note that this process only shifts the image. No distortions, rotation or scaling.


A better option may be to use the new TaylorRegister process that is currently under development. It may achieve very good subpixel accuracy with moderate translations, and small rotations (under 1º). At this point we are testing the algorithm with real data, so maybe yours is a good candidate for it. If you want to share a couple of images, it would be very nice to try the algorithm on them.

[pfile]

related to this, i took some very low-tech images of the eclipse using a 200mm lens, and a lens cap with a hole drilled in it; the hole was covered by the film from a cheap pair of eclipse glasses.

i did not use my GEM and so i had to reposition the camera several times during the eclipse - the camera was mounted on a tripod with a ball-head. because of this i introduced small rotations each time i repositioned - about 1-2 degrees each time.

i loaded the pictures into Lightroom 4 and cropped them in preparation for registax, but that did not work out well. i switched to using FFTAlignment on the set of images and it worked beautifully.

so then i decided to fix the rotations (in lightroom) and try to re-register them with FFTAlignment. perhaps 20% of them now failed to register.

i'm guessing that the rotations must have introduced some high-frequency artifacts which confused FFTAlignment.


Carlos, i'd like to try your modules but am running OSX. in general it seems like the lack of a way to cross-compile these modules is a problem for those of us that use a different platform than the developer...

[Carlos Milovic]

I may send you the source code... although, I've being having troubles with recent real world data, so I think I'm gonna introduce several changes to the current code to make it more robust. Due a extreme overload of work here at the university, I guess that I'll be unable to do that until 2 o 3 weeks from now.

[pfile]

okay no problem. my project is on the slow boat - i just took flats today, 2 weeks after the fact.

[TrickyDicky]

A better option may be to use the new TaylorRegister process that is currently under development. It may achieve very good subpixel accuracy with moderate translations, and small rotations (under 1º). At this point we are testing the algorithm with real data, so maybe yours is a good candidate for it. If you want to share a couple of images, it would be very nice to try the algorithm on them.

OK, Why not, ....
I have 12 images [in JPG] of roughly 4MB each. Will upload them somewhere and send you links. Or do you prefer an e-mail with attachments?

[TrickyDicky]

OK. Here is the link:
http://www.4shared.com/rar/A_8f2mO3/Saturn-14-Apr-2012.html
 
#TD

[Carlos Milovic]

Thank you very much! Tough image for those algorithms... both failed to align the images sorry! I think I'll need a lot of time in the drawing table improving them. At least, Taylor registration was a bit better...

[TrickyDicky]

So, do you think will there be any positive outcome of this exercise?

[Carlos Milovic]

For this kind of images, yes. A centroid estimation will help a lot to force the convergence. I can implement that quite fast.

[pfile]

hopefully it will be configurable as if you estimate centroids on eclipse images the registration gets weird. this is what happens in registax if you select "center of gravity" which i assume is the same as a centroid. maybe not.