From Rho to Antares mosaic

[gvanhau]

Hello
This is a three pane mosaic (2 horizontal 1 vertical) of this beautifull area in the sky.

I had a lot of work in aligning the three images using dynamic align. (star align did not work properly, there where always patches with double stars, etc.).  On the third image I had to point more than 500 star pairs in order to get this reasonable result...

The equipment used is a WO 98 FLT APO telescope and a modified Canon 5DMKII DSLR on top of an old CG5 mount.

Images and mosaic fully procesed in PixInsight.

Here med and full res images:
https://www.flickr.com/photos/astro_gvanhau/14645399061/sizes/h/
https://www.flickr.com/photos/astro_gvanhau/14645399061/sizes/o/

Comments and critics welcome.

Geert

[Juan Conejero]

star align did not work properly, there where always patches with double stars, etc.

That should never happen with a mosaic like this one, if StarAlignment's local distortion correction feature is used properly. Can you upload the (linear) mosaic frames, so I can take a look?

[gvanhau]

Hello Juan
Thanks for your concern.
I am uploading two of the three images now. I see it will take some time since each image is 247MB.
I will send you a PM when ready.
I can also tell you that I tried to align the images against a star field  generated with CSG script without success.

Regards
Geert

[Andres.Pozo]

I can also tell you that I tried to align the images against a star field  generated with CSG script without success.

Hello Geert,

For widefield mosaics I think the best approach (IMHO) is using ImageSolver+MosaicByCoordinates. It has the advantages of that it doesn't require overlap of the images and the result has coordinate keywords so you can annotate it. It allows also to change the projection if the native gnomonic projection is not suitable for the result.

If you are uploading the images, I would be interested in using them for testing the scripts. Could you send me the link?
 

[Juan Conejero]

Hi Geert,

I completely agree with Andrés, his ImageSolver and MosaicByCoordinates scripts are without doubt the best tools to generate wide field mosaics like this one in PixInsight, especially for mosaics with more than two frames.

Besides that, I am interested in trying out StarAlignment with your images because I am always interested in improving my arbitrary distortion correction algorithm and its implementation. Non-rigid image registration is one of my favorite R&D areas.

[Andres.Pozo]

Hi Geert,

your optical system has geometric distortions that make difficult to align or solve the images. I have generated a distortion model (attached) that helps both StarAlignment and ImageSolver. There is attached also a visualization of the distortions.

I have been able to register your images after selecting this distortion model in StarAlignment.

Also, using the distortion model and polynomials of 5th degree the images get solved really well. After this, MosaicByCoordinates can generate a very accurate mosaic.

[gvanhau]

Besides that, I am interested in trying out StarAlignment with your images because I am always interested in improving my arbitrary distortion correction algorithm and its implementation. Non-rigid image registration is one of my favorite R&D areas.

Thank you Juan. You received my images didn't you? if not, I will send them to you again. Anyway, the results of Andres may explain why it is so difficult to align them well.

Hi Geert,

your optical system has geometric distortions that make difficult to align or solve the images. I have generated a distortion model (attached) that helps both StarAlignment and ImageSolver. There is attached also a visualization of the distortions.

I have been able to register your images after selecting this distortion model in StarAlignment.

Also, using the distortion model and polynomials of 5th degree the images get solved really well. After this, MosaicByCoordinates can generate a very accurate mosaic.

Thanks a lot Andres.
I agree is a very strange distortion pattern. 
I didn't know I could select a distortion model in Star alignment...  I will try all this stuff asp.

Geert

[Juan Conejero]

Hi Geert,

StarAlignment generates your mosaic without any problems. Enable local distortion correction, select surface splines as the registration model, and enable frame adaptation. Select one of the mosaic modes (separate mosaic frames in my example), and leave everything else with default parameter values:

Click for full size image

The result is very accurate on the whole intersection area. One of the mosaic frames is rotated more than 90 degrees clockwise. This wastes some computing time to find the intersection between both frames. You can save that time by rotating the frame in question 90 degrees counter-clockwise before StarAlignment (with FastRotation).

As Andrés has shown these optics have a rather complex distortion pattern. However, StarAlignment's local distortion correction algorithm is able to correct it without problems. The algorithm has used 20 iterations to achieve a residual of 0.007 pixels in the registration model. You may want to set  the distortion residual parameter to a more realistic value of 0.01 px, instead of the default 0.005 px.

[gvanhau]

Hello Juan

Thanks for taking your time.

The difference between your settings and mine was the Resgistration Model. I left de default Projective Transofrmation.
Now, using 2d SP-linles, although the first two images aligned well, when combinig the resulting partial mosaic image with the third image,the result was not satisfactory:
I got two different results depending if I selected register union-mosaic or register union-separate.
The first option registrated a "aceptable" mosaic, but with visible seams. Register union-separate generated two images, that when overlayed showed visible unmatched stars. Also, if I overlay any of the two separate mosaic images over the full mosiac image, there are lots of unmatched stars.

I can send you the third image if you want to, but in that case, I suggest to move this thread somewhere else....

I attached the results in reduced jpg's, but they show what I'm saying.

Regards
Geert

[Andres.Pozo]

When aligning more than two images with significant distortions usually is necessary to use a common reference frame covering the whole field. You can generate a synthetic image using StarGenerator or CatalogStarGenerator. You can also download an image from DSS using DSSImageDownloader.

Then you can align your three images against the common reference image using the mode "Register/Union - Separate". After this you can combine the with PixelMath or GradientMergeMosaic.

[gvanhau]

Hello Andres

I tried this last night:

- Generated a tar field image using star gen script with a center defined near the center of my mosaic and about the size of my mosaic  and stars up to mag 14 (more crashed the script).
- Aligned first two images using the distortion pattern you gave me and checking undistorted reference, also checking  distortion correction, 2d splines and 40 iterations one image converged after 19 itereations, the other did no converge after 40 iterations. When superposing these results, no misalignment was visible.
- Aligned my third image using a distortion pattern generated using the  image resolve script. (this pattern was different because I used a different flattener when I took this image. The alignment converged after 21 iterations.
- Generated the mosaic using gradient merge mosaic.
- The result  was very good, but not perfect: there was a mis aligned star field between images 1 and 3 (I would have espected a mis alignment between image 2 and any of the other two).

So, this aproach is very good,  but my result using dynamic align is better.

I dont have the images here at my office, but if you want I can post them this evening.


Regards
Geert

[Andres.Pozo]

I already have the two that you uploaded to Endor. I would need the third.

FYI, the test that I did with that two images using MosaicByCoordinates, produced "perfect" alignment.

I am going to be extremely busy for the next days so I probably won't be able to do more tests until the next week.