Aligning ultra wide field images

[mstriebeck]

This week I took lots of images with my DSLR at 14mm to catch Perseids.

I am trying to create a combined image where I have all the star trails but have a really hard time registering them (the camera was fixed on a tripod).

This is my base image: https://drive.google.com/file/d/0B0KcEqw-v71VV0NaZWpYQmZmNVk/view?usp=sharing

This another image with a meteor trail that I can't register:
https://drive.google.com/file/d/0B0KcEqw-v71VVnQxVnRxMHFQQjA/view?usp=sharing

And here is one that does register:
https://drive.google.com/file/d/0B0KcEqw-v71VYlRPazZNWUFOYWM/view?usp=sharing

What seems to be the reason is that the distortion of the 14mm lens makes it hard for Pixinsight to match stars. I tried using "2-D Surface Splines" as the registration model, but that didn't work either.

Last night I tried using DynamicAlignment (which would be a real pain in the ... to do!) but couldn't figure out how to indicate on the second image where the stars from the first image match.

Any ideas are welcome - thanks!

     Mark

[msmythers]

Mark

I was able to get the images some what aligned with DynamicAlignment. I use 460 points for one image and 683 for the other. I probably should have used more.

The problem as I see it is one you have the different distortions of the lens and then you have the field rotation. As the field rotates the lens distortions have different stars in those areas. If you can map the lens distortions first and apply that correction to each image I would expect alignment would be easier. I don't know how to do that within PI but I think the place to start might be the Image Plate Solver script or the Structure detection in StarAlignment.


Mike

[Juan Conejero]

Hi Mark,

The StarAlignment tool has not been designed to solve these problems, so it is useless in this case. Your best choice is probably the ImageSolver and AlignByCoordinates scripts, although I am not sure if such huge distortions will be manageable by these scripts. DynamicAlignment can align these images without problems with very high accuracy, but requires manual work---less work than what one can think at first glance, but if you have lots of images...

As Mike has pointed out, the best way to align these images is by computing a distortion model first. You can use the ManualImageSolver script to generate it, starting from a  DynamicAlignment icon. This script is also fully documented.

[mstriebeck]

Thanks a lot for looking into this. Looks like I should investigate more into the DynamicAlignment tool. I thought I have set (and correct) all alignment points manually. Is there somewhere a good tutorial?

Thanks
     Mark

[msmythers]

Mark

The way I set the points is I start with some very large stars in the center of the reference image or stars near a DSO. Those are easy to recognize. I have the 2nd image zoomed out in case the the first few point aren't located close to where you start. Once I get 10 to 20 points around that central area I move towards a side. I then work my way around the image in one direction, clockwise or counterclockwise. This allows me to build the points with the least amount of false locations. And if a point is not correct it is generally very close and easy to move over the correct star. Once I have gone all the way around the image I can fill in the rest of the image if needed. With compound lens distortions you might have to fill in some tight areas.

That method works every time for me. Just make sure you try to get to the edge of the reference image if possible.


Mike