The three first images (1-1) fisheye spherical, (1-2) cylindrical and (1-3) "straight" rectilinear
are a proof that none of the "regular" projection modes can result in a printable image
when the horizontal field of view is larger than 120°.
(1-1)
Source image: full-frame fisheye to be adjusted to give every of the following images - 130° horizontal FOV
(1-2)
Cylindrical - 130° horizontal FOV - right part is OK, car is much better than it's in the fisheye version (1-1), but...
In the next two images (1-4) and (1-5) the previous rectilinear one (1-3) has been "squeezed"
(the red and blue lines are located in the middle of the unmodified region.)
(1-4)
Squeezed rectilinear - 130° horizontalal FOV - car is fun but..
(1-5)
.. easy to replace with a rectilinear one (over unchanged background)
The next three images (1-6), (1-7) and (1-8) are an illustration of the method explained
by L. Zelnik-Manor in his paper Squaring the Circle in Panoramas (see references)
where two or more panorama rectilinear sections are joined and each of the section
uses a different projection plane. This method avoids the exagerated stretching
which would occur at both sides of an usual "single section" rectilinear panorama.
(1-6)
Double rectilinear - 130° horizontal FOV
(1-7)
Double rectilinear, a different plane was used for the left part projection,
right part is identical - 130° horizont FOV
(1-8)
Same as above rectilinear left part is stitched with a cylindrical right part
(at this reduced image size the difference is quite subtle) - 130° horizontal FOV
In each of the three previous examples, the stitch between the two projections is located
on the right side of the large blue building wall wich supports a series of logos.
A DIFFICULT EXAMPLE
Built in 1964 by Anger et Pucinelli in Grenoble, France, those three 30 levels white towers
are not very easy to include together in a single view.
The image bellow (Minds-Eye-Viewer, unknown projection mode) gives a good idea
of their disposition along a straight avenue, but!!!
The 210° cylindric projection view (opposite side) distort them in such a way that they are
no more identifiable (note the shape of electric power cables!).
(2-1) Cylindrical version - 210°
Using a viewer is fine (bellow), but this don't work if one want a print:
(2-2) Equirectangular displayed as rectilinear (SPi-V) - 210° Please use the mouse to scroll the view, Ctrl and Shift to zoom
The left tower, the large tree and the right part were the three projection planes I first used:
this was not enough, a projection plane for each of the towers in the right part is much better.
Having no easy to follow rules to decide of the amount and the location of projection planes
will certainly appear to most photographers as a too difficult situation but if you compare
the best of above versions with the hideous cylindric version and remember this is an unusual image
where you can see both ends of the street (FOV: 210°) and that architecture is not regarded as
the easiest part of photography ...
Note that the only possible way to solve the cables problem is to delete them (and, please, ignore the
stitching errors I didn't have the patience to deal with.)
(2-3) Place the mouse cursor on this 4 projection planes
vertically-squeezed-rectilinear version to display an "only 3 projection planes" version - horizontal FOV 210°.
Using PTAssembler to produce multiplane panoramas
(3-1)
Rectilinear: the left part is not acceptable, stretching is less visible on the right part as this is the dark side of the street
- 133° horizontal FOV
(3-2)
Rectilinear left part - 51° horizontal FOV
(3-3)
Rectilinear center part - 51° horizontal FOV
(3-4)
Rectilinear right part - 51° horizontal FOV
(3-5) Place the mouse cursor on this 3 projection planes rectilinear version
with vertically squeezed tower to display an unsqueezed tower version - horizontal FOV 130°
To produce this 3-plane rectilinear, I fisrt optimized a standard rectilinear 133° image (3-1),
using t1 lines to get the correct pitch and roll values.
Then I changed the Step 5 Horiz. FOV value to 51° (somwhat higher than 133/3 = 44°) used
Reference Point Picker screen to guess the yaw value for each part (but used Cancel rather than OK button)
and for each of those parts changed the yaw value on Step 1.
The most difficult was to find correct Width and Height values on Screen 5, but when a correct value was
found for one of the part, I reused the same values for the two other ones (a more straightforward
workflow is possible if only the yaw is changed betwen parts creation: larger files, but easiest.)
The final steps were done in Photoshop using layer masks (image bottom part should be cropped
or a fourth part used to avoid rails having this unusal shape):
