Lens distortion. Mosaic construction problem

[StasV]

Hello!
Please help

I have a problem with mosaic construction
Here are two frame of the future mosaic
Canon 200mm f/2.8 + STL11k   each

This lens have distortion at the corners. And I can not combine  two frames correctly.
Look to pictures below.
Frames after the stars alignment look “very rectangular”. As if the PI did not try to correct the lens distortion.

I tried to work in different ways: With and without Reference star field. I used Projective transformation mode and 2-D surface spines.   In general, the result is the same. Distortion does not allow to combine the images

Details:
A1 center Ra 05 12 13 Dec +10 04 06
A2 center Ra 05 12 31 Dec +04 03 30
Mosaic center Ra 05 12 22 Dec +07 03 48

A1.fit  A2.fit and reference star field here http://depositfiles.com/files/1a7mzgc9c

[Juan Conejero]

Hi,

StarAlignment cannot correct so large distortions. Increasing RANSAC tolerance and forcing it to maximize the area covered by star pair matches helps a lot (see attached screenshot), but does not solve the problem completely with your images.

Your best option to build this mosaic is DynamicAlignment. use your "stars" image as reference (select it as #1 in DynamicAlignment) and one of the mosaic frames as the target image. Define some 50 alignment stars for each frame, well distributed across the whole image. Make sure you cover well the overlapping area. It requires some manual work, but looks much more difficult than it actually is. Once you have defined a few star pairs, DynamicAlignment finds the rest automatically. Watch the Ex and Ey values; they tell you the differences between the coordinates predicted by the model and the coordinates of each new star. With a little practice you can have your mosaic ready in half an hour.

Another option: Build the mosaic with StarAlignment, using the parameters shown on the screenshot, but with the "Register/Union Separate" mosaic mode and frame adaptation enabled. Then apply DynamicAlignment to the aligned mosaic frames. This will fix the residual errors due to distortion, and is much easier and faster than aligning against the synthetic star field.

[StasV]

Thank You Juan!
I did a few tests started with your method #2

First way
1. Star Alignment   “stars”-reference. “A1”-target. “A1_registered” – result 
2. Star Alignment   “stars”-reference. “A2”-target. “A2_registered” – result
3. Dynamic Alignment  “stars”-reference. “A1_registered”-target. “A1_registered2” – result (about 150 ref stars)
4. Dynamic Alignment  “stars”-reference. “A2_registered”-target. “A2_registered2” – result (about 150 ref stars)

You can see the  A1_registered2+A2_registered2  at the picture below (pic1).
The result is not perfect. There two arias with bad alignment.

Second way
1. - 4. the same with First way
5. Dynamic Alignment  “A1_registered2”-reference. “A2_registered2”-target. “A2_registered3” – result

A1_registered2+A2_registered3 is shown at the pic2 
It is mach more better! May be excellent.

In my opinion the cause of Fist way problem in dynamic alignment between ideal stars field and real image. For example, the is small star right near big star. Real image contain both stars. But star field only big star due to magnitude limitation. Centroids well be calculated different ways.

Unfortunately, I found, 2-D Splines are wary sensitive to wrong points. Only one bad star pair  can launch oscillations far from  this bad point.

[StasV]

pic2

[StasV]

I am working on 8-frames mosaic now. It is analogical to  make same operation 8 times. What well be with 20- frames…… 
It would be nice to see Distortion Correction Module as separate module or a part of star alignment module.
Idea is clear:  Make distortion model once as accurate as possible and use it every time after by one click.

[starhopper]

Hello StasV,
thanks for posting your problem. I have a 2 frame mosaic from namibia made with my STL11000 and a Canon EF200mm Lens.  After working two evenings without any reasonable result I checked the forum.
Unfortunately, I had to export as tif and stack it with PS Photomerge, it was a 15s task and the result was much better.
A distortion module within star alignment would be nice.