in theory, yes. This would require a sensor that is absolutely not tilted/swinged/rotated (at least less than 6.8 microns on the said P45) and at the same time a camera with a perpendicular alignement on every axis within the tolerances of the sensor's pixel pitch. The camera would have to hold these tolerances also at large movements. I don't think that such a camera exists.
That's correct, but in practice the differences are small, and registration to less than 1 pixel difference is feasible. A misalignment by a fraction of a pixel will be very hard to detect. Sensor rotation requires translation in 2 directions to eliminate most of it. By doing a manual transition mask in the overlap area, the residual error becomes even smaller.
Too, it would require a tripod and tripod head that moves less than the sensor's pixel pitch when you operate the camera (i.e. when you adjust the shift for the stitching).
Good technique is assumed, but you are correct in mentioning it.
Finally it would also require a motif without any movement (clouds, trees... whatever).
Obviously, that applies to all approaches that require multple exposures.
That said I have infact been stitching some images manually and it wasn't really a problem.
Indeed, my experience as well. When the overlap between image tiles is large enough then a lot can be covered up, including some subject motion.
Cheers,
Bart
