Problem with Merging simple Mosaic

[fneyer]

Hi PixInsight Experts,

I have problems to merge a two-frame mosaic with 'high accuracy'. What I want to do is the following:
- Generate starfield with StarGenerator
- Align both mosaic parts to the starfield
- Merge both mosaic parts to a single image


Now whatever interpolation algorithm I use there are always distortion errors (in the order of 2-6 pixels). Since my images were taken with a TEC140 APO, its dedicated FF, and an STL11k, the field only has minor distortions (not as perfect as with the FSQ106 but better than with any other scope I used before). So my question is if anyone can manage to get this done with PixInsight.

If you would like to try, here are the files:
http://www.starpointing.com/temp/mosaicBottom.fit
http://www.starpointing.com/temp/mosaicTop.fit
http://www.starpointing.com/temp/Starfield.fit

I'm very much interested in what you get and how accurate it is!
Any help is very much appreciated!

Thanks and happy Eastern,
Fabian

[ruediger]

- Generate starfield with StarGenerator
- Align both mosaic parts to the starfield
- Merge both mosaic parts to a single image

Is the mosaic going to become larger than only the two parts? If not, a starfield is not necessary, you can use StarAlignment / Register /Union - Separate, see screenshot. Not shown is the interpolation setting: 2D-surface splines, Auto, 0.30.

After tweaking the parameters, I got no visible errors when I checked the difference of the aligned images and the average of them. Of course you have to linear fit them and use the Gradient Domain Tools to really build the mosaic.

I'm not sure, if these parameters will work with the starfield also, but I made just a quick test on my netbook which is not the fastest when aligning images

Happy easterweekend also from me!

Rüdiger

[fneyer]

Wow, what a reaction time.. Thanks a lot, Rüdiger!

Indeed, I could have used only the two parts without the starfield. However, my idea is to limit the number of image transformations when building a mosaic. Since each transformation reduces the data quality, I want to build a reference image which is suitable as a reference for all parts of the mosaic. Now if you would create a reference image without the starfield, i.e. just from my two images, the distortion easily becomes too large and the alignment for the subsequent images will most probably fail (because distortion adds up and gets worse the larger the mosaic will be).

I will try the settings you used tomorrow and see how it goes!

Thanks for your help!
Fabian

[Juan Conejero]

Hi Fabian,

You can build huge mosaics (e.g. 16 frames) without a reference star field. The number of interpolations is always the same: just one interpolation for each mosaic frame except the first one, which is never interpolated. Bear in mind that StarAlignment only interpolates the frame that you add to the mosaic at each step; the reference image is only copied in place.

With significant distortion a synthetic star field won't work. You need to align with stars at the overlapped area between each adjacent pair of frames:

Click for full-size image

Note the increased RANSAC tolerance parameter (6 pixels), which improves accuracy under high distortion. This screenshot shows the seam between both frames with the generated registration mask active on the mosaic image:

Click for full-size image

Another important option for accuracy is surface spline coordinate interpolation, as Rüdiger has pointed out. By the way, no previews are necessary to build this mosaic; StarAlignment can find the intersection area automatically.

Finally, you probably will want to use Georg's GradientMergeMosaic tool. This requires building the mosaic in register/union - separate mode.

Hope this helps.

[fneyer]

Hi Juan,

Thanks for your help!

My point in creating a reference image is to reduce the total number of interpolations. Usually you align all images of each mosaic part to one-another. Afterwards you would stitch those parts togeter. So for all but one mosaic part, the number of interpolations is 2.
If I now could align all images directly to a reference representing the final star positions of my whole mosaic, I would have to make only 1 interpolation. Of course you also can create a reference image with two shots of each part of the mosaic (like you did). The problem I see in this is distortion: For larger mosaics, slightly distorted single frames may produce a reference with a net distortion that is much larger. If you could manage to use the star catalogue as reference, you could get rid of that. Does it make sense to you?

Fabian

[ruediger]

The problem I see in this is distortion: For larger mosaics, slightly distorted single frames may produce a reference with a net distortion that is much larger. If you could manage to use the star catalogue as reference, you could get rid of that. Does it make sense to you?

This is partly discussed though not finished here: /forum/index.php?topic=4071.msg28343#msg28343.
I think there is not much missing to use the coordinates from plate solved images for mosaic building on star fields. Especially it helps if you are going into the range of camera widefield lenses. Also you could start placing arbitrarily pictures onto a grid even if they don't overlap at all, showing you the gaps to fill with additional pics.

Rüdiger

[fneyer]

Exactly, Rüdiger!
And you would save one image interpolation which gives you better data quality (even though just a little bit)..

Fabian

[fneyer]

Hi Rüdiger and Juan,

Now I tryed to align my two mosaic parts to the starfield using the parameters you estimated. The result is attached.
Also when I try to use just the two mosaic parts (combined to a mosaic) as a reference to align all other images I get a very similar result. So neither technique seams to work. I guess the distortions are still too large to be reduced in a single transformation.

Fabian

[Juan Conejero]

How about this:

- Generate a working mosaic with StarAlignment following the sequence that minimizes cumulative errors. For example, to generate a mosaic with 9 panels, make 3 submosaics, one for each row of three images, and then the whole working mosaic taking the central row as reference. Or do it by columns, if more appropriate.

- Now generate your final mosaic using the working mosaic from the previous step as reference. In this way you'll only interpolate each image once. If your mosaic is very large (> 12 frames), perhaps you'll have to help SA defining previews roughly covering the overlapped area for each frame, although the automatic intersection computation routine works nicely in most cases. If you want to use GradientMergeMosaic you'll have to generate separate aligned mosaic frames.

We have used this technique to build very large mosaics with excellent results.

[fneyer]

Yes, that is a possibility! Thanks Juan for your thoughts!
However, in this case you use your images for the working mosaic which, eventually, are distorted. I was more like thinking to reduce these 'minor' distortions using the starfield from the SG tool to build an 'undistorted' working mosaic.

Just out of curiosity: What telescope did you use for building the mosaics you mentioned?

Best regards,
Fabian

[bosch]

***

Another important option for accuracy is surface spline coordinate interpolation, as Rüdiger has pointed out. By the way, no previews are necessary to build this mosaic; StarAlignment can find the intersection area automatically.

***

Gracias Juan,
llevaba todo el fin de semana loco modificando parámetros y me era imposible unir más de 4 de las 16 teselas de mi mosaico, hasta que finalmente he escogido "2D-Surface Splines" en Registration model.