[Solved] Pattern after registration?

The pattern looks very regular indeed. What settings did you use in APP?

Hi,

I usw the default settings. As they are after starting APP.

@daniel_n Hi Daniel. Do you get the pattern as well if you only integrate the lights, i.e. not load any calibration files?

Hi,

just tested. APP started. Loaded two light frames and clicked directly on register. One light has the exact same pattern.

I've done a lot in APP and this has never happened before.

I have upload the files in the File Station / pattern after registration

@daniel_n Thanks for the upload. I loaded the two unprocessed lights into APP 1.083.2 (macOS) and registered them. The second light is taken as reference and the first is registered against it. After registration is has been rotated over less than 5 degrees so the two images match up. This has introduced the pattern. I'll discuss with Mabula and will get back to you.

Dear Daniel,

I have just checked your registration phenomenon.

What you are seeing here is interference of your sensor characteristics with the resampling algorithm being applied. You can check this with the l-registered image viewer after registration.

So when registration has finished, go to 6) integrate and change the pixel interpolation filter and enable/disable no under/overshoot, double click the registered file (or reload the l-c-registered image viewer by reselecting it from the dropdownbox above the image viewer while the registered file is selected below). Depending on the registration parameters and the chosen filter, interference patterns can occur ! The under/overshoot protection plays a role as well.

The problem seems to be worst with the bilinear filter which really indicates that an interference pattern is created of your sensor characteristics with the registration parameters...

To make especially interesting, check what happens if you use the OpenGL versus the CPU image viewer. Clearly, even OpenGL texturing plays a role here, because the CPU image viewer shows it much less! If you zoom in to 100% the pattern disappears, I can not see it...

The problem seems to be gone if you use Cubic B-Spline pixel interpolation. Even with OpenGL the pattern is gone.

And... you can also try drizzle to reconstruct data:

in 6) Integrate, at the bottom, select drizzle, scale 1.0, droplet size 1.0 with tophat kernel and check your registration with l-c-registered, then the problem beccomes very clear ! So it is definitely not an APP issue, but rather a freak occurence of sensor characteristics creating interference when registration is applied... an oscillation seems to occur and OpenGL texturing enhances this, allthough the actual data has it less as seen with the CPU image viewer.

If you drizzle and increase the droplets to about 2.0, the pattern is gone as well, which is consistent with Cubic-B-Spline because we are starting to blur the data a little bit then...

It will be very interesting to see if applying proper dark frames in calibration makes this problem go or if it has no influence. Have you tried to calibrate the data with darks and flats?

Mabula

@mabula-admin 

Thanks for the answer.
Yes you are right. Very interesting. When I switch to CPU this pattern is gone.
Also when I zoom into the 100% view, it is no longer visible.
I was just a bit surprised because I had never seen this in APP before.
I have calibrated the lights with the darks and flats. See 1st post. It can also be seen here.

And yes, it's definitely not a problem of APP. It's still very interesting how this can come about visually.

Regards,
Daniel

You are most welcome Daniel @daniel_n 😉 !