Image registration comparison

[Avdhoeven]

Recently I noticed that dependent on the software that I used for image registration I got different starsizes. This triggered me to do a simple test. I compared six different packages. I took a dataset of good h-alpha data that I had and registered the images in the different packages, saving them for further use.

Next I used CCD-stack for normalization and combine (mean) without any pixel rejection involved. Then I cropped all images to roughly the same area and loaded everything in ccd-inspector to check the fwhm values.

As the only difference between the stacks was the alignment used this gives an indication of the quality of the procedures involved.

The toplist shows the fwhm from the different stacks, the bottom list shows the data of the seperate images used in the stack.

I have drawn my conclusions which software I will use from now on for aligning my images

I'm not involved with any of the packages and just wanted to make a clean comparison.

[jnc71106]

so which is the better overall in your opinion ?

[davew]

Just out of interest and never even thinking about this before, I did a test of my own using DSS and Maxim. I stacked a couple of different ways. DSS was the clear winner in the FWHM stakes !! I never would have believed it. However, Maxim won the noise test.

The only way I would differ from your methods would be to calibrate and stack, as you normally would, in each program. Not bits here and there. Unless that's the way you work.

Nice thread idea,

Dave.

[Avdhoeven]

I found out that the interpolation scheme makes a huge difference as some of the procedures use a build-in sharpening. In the end almost all are pretty close...

[jnc71106]

I'm going to try a few images as well. I will throw astroart into the mix as well

[ollypenrice]

Oh, damn, I stack in Astro Art and you didn't try that! I know Harry Page has found that Pix wins on FWHM and I'm not surprized to see Registar up there at the top, though it doesn't calibrate so I use it for colour channel alignment where it beats Astro Art convincingly sometimes. I also use it for mosaaics and high resolution overlays.

But the other thing is noise and the quality of faltfield correction, which I find seem to be inter-connected and quite variable.

Good post though. Thanks.

Olly

[jnc71106]

if the original poster wants to send me the image files via dropbox or something then i will happily register them and send them back for him to analyse

[dph1nm]

What is the native pixel scale? How well you are sampling the PSF must have some effect on the results.

NigelM

[davew]

I found out that the interpolation scheme makes a huge difference as some of the procedures use a build-in sharpening. In the end almost all are pretty close...

I've decided this quote is the most sensible for me. Who knows what each program does behind the scenes ? Would the worst in the FWHM stakes be the least meddlesome ? If a corresponding deconvolution routine was run on the worst would it equal or better the best ?

So, as I've been told in the past, you get what you pay for. Use the one that cost you the most

Dave.

[IanL]

Who knows what each program does behind the scenes ? Would the worst in the FWHM stakes be the least meddlesome ?

In the case of PixInsight, you can find out exactly what it does behind the scenes here:

http://pixinsight.com/doc/tools/StarAlignment/StarAlignment.html

I don't know about the other applications being tested, but there is no sharpening or deconvolution applied during image registration in PI. I suspect the most critical issue is choice of pixel interpolation algorithm; there are three choices in PI and an 'auto' mode which chooses one of the three based on how much scaling is needed for a given image. I'm assuming the results above were obtained with the default PI alignment settings 'out of the box', so mileage might vary depending on the specific images.

[RobH]

Interesting info.....thanks for posting this.

Of course, as soon as you start stretching, no matter how careful you are, it all goes out the window anyway!

[jnc71106]

I've also run the same test but threw astroart into the mix. The results were similar with PI coming out best for FWHM but astroart and PI were almost identical on everything else

DSS, Maxim and CCDStack were not quite as good

I must say I personally prefer PI now having been a fan of astroart before