I have a series of 180 subframes, split into two approximately equal size groups that overlap only about 80% in each dimension. The two groups are both at the same scale, just offset in the targeting. Let's call these group A and group B, where group A appears first in the list of images to register and integrate.
Suppose I would like to preserve all of the data, such that the output image is larger than the individual subframes, with a central portion to which all subframes contribute, bands around the edges where either group A or group B contribute, and two black regions in corners where no data are available. In this case there are two apparent problems:
1. The output image dimensions and coordinate system are entirely determined by group A. Part of the group B extent will lie outside this range and be discarded.
2. In areas of the image that are covered only by group A, I suspect pixel rejection algorithms and other statistical calculations may incorrect. I can see this in what seems like too much signal being rejected in these regions. This could occur if missing pixels in the pixel stack are viewed as correct black pixels and are being included in the statistical calculations. This would bias many of the statistics including mean, median and standard deviation, likely causing a sigma-based algorithm to find more true signal to be above mean+k*sigma. It would also affect normalization, SNR calculation, etc., thus potentially having an adverse effect on the central portion of the integrated image, where all subframes contribute.
I'm describing an extreme case but if problem #2 is real it would be present to some extent even with a relatively well-aligned set of subframes.
To solve #2 (the more serious problem, in my view), it seems like the registration algorithm would have to mark part of the registered image invalid, rather than just set it to zero. Then statistical calculations would have to ignore those invalid portions.
John
