Hi @supernov,
Okay, I have been testing your data. My findings:
if we don’t use distortion correction, the registration RMS is already very good considering this is debayered H-alpha data.
Using distortion correction with the “same camera and optics” option turned on, gives a slight improvement in precision.
Using the distortion correction with the “same camera and optics” option turned off, triggers an exception in the multiple view regression algorithm. It keeps finding better solutions, but the solution becomes “unstable” due to the fact that all images have almost exactly the same field of view.
For distortion correction to work optimally, it’s better to dither more aggresively, the distortion correction will be of better quality and this exception won’t be triggered. But I need to catch this exception properly off course and prevent it from happening.
If only one of the frames to be registered has a field of view that is different by for instance 50 pixels, this exception won’t occur. So I think i can prevent this from happening by checking the homography parameters (i.e. the coordiante transformations with scale, rotation, perspective, translations between the frames) and check if all frames have parameters relative to the reference which are really similar to the identity matrix… if so. The complex model shouldn’t run.
Another solution would be to let the complex model run, but detect the solution from becoming unstable.. which would also be based on the homography parameters.
To be continued…