Do parallax corrected subframes *need* to be warped when pano stitching? That is, can a pano comprised of images revolved around the nodal point be successfully stitched by repositioning only or do they need to be warped to some projection?
Hi,
The short answer is, yes. The longer answer is, most likely yes, unless you want to produce output with flat panels that are arranged in a spherical orientation.
Whenever you want to project an image on a surface, the projection will distort the image. Take for instance the left most image of a single row of three images. The camera, presumably with a rectilinear lens and a flat sensor, looked at that part of the scene at an angle, different from the center image. That angle distorted the image
relative to the center image, and thus needs reprojection if you want it to not produce abrupt kinks in straight lines and edges. Same for the right most image but in opposite direction.
Add to this that no lens is absolutely free of distortions, and even perfect perpendicular alignment of the optical axis on the middle of the sensor is within a small margin of error. The stitching software can take care of all of those variables.
My concern is pixel level smearing due to warping.
Dedicated stitching software typically offers a choice of resampling algorithms, and they are usually a lot better than Photoshop's Bicubic in maintaining micro-contrast, even with sub-pixel shifts. Shooting panoramas, is also usually done with a longer focal length than one would otherwise use to get the FOV. Longer focal lenses are usually better corrected, especially when dealing with edge and corner detail, and they offer higher resolution due to a larger magnification on sensor. So you build in additional resolution which can partly be used by resampling.
In addition, many panoramas are so large that they can be down-sampled without sacrificing output resolution, or at least they need less upsampling, which all benefits final resolution.
Cheers,
Bart