Plagues on proximity panography

This is a complement another earlier technically detailed and closely related article

Subject

I have done a few panoramas which the subject was very close to the fisheye lens.

Some have become famous. Some are still to be shown at the date of this writing.

I have stumbled against specific difficulties while doing so and I wanted to share my experience in this very exotic photographic field. This problems are at least:

Illustrations

Experimental setup:

Two metallic grids are put about 6 cm apart and the fisheye lens lies on the upper grid.

Canon EOS 5D Camera, Tokina 10-17mm fisheye lens.

One shot is made having a central rod of the upper grid centered in front of the lens:

Image A

f= 11 mm; f/22; 1600 ISO; 1/5 sec

Another shot having the lens axis just between two of the rods:

Image B

f= 11 mm; f/22; 1600 ISO; 1/5 sec

First problem: parallax errors

This is due to a specific property of the fisheye lens: the entrance pupil is seen at a position on the optical main axis of the lens at a varying position depending on the angle which the lens longitudinal axis is seen by the observer. In other word the NPP (no-parallax-point) shifts along the axis in function of the angle that the line joining the point of interest and the entrance pupil makes with the lens longitudinal axis.

On most of the fisheyes that are intended to be mounted on (D) SLR, the NPP locations range from the front glass surface of the front lens to about 18-20 mm behind on the optical axis. This correspond to ~97 degrees and ~0 degree angle respectively for the lens used here (i.e. Tokina 10-17 mm @ 10 mm).

With some slight variation this is true for any fisheye lens including of course the Sigma 8 mm f4 for instance.

This shift is of no consequence if the point (or object) of interest is quite distant, but it has a strong influence on the position of the output image corresponding picture element (aka pixel) location on the digital sensor when the distance becomes small (e.g. few centimeters or few millimeters to the entrance pupil).

  1. This implies that unavoidable parallax errors shall exist on the seams between two images to be stitched.
  2. This also implies that strange shape shall be given (and be noticeable on the defished rectilinear output image) to straight edges or slender objects close to the fisheye lens, depending also on the position of this edge relative to the diameter axis of the lens:

Rectilinear outputs:

The following two rectlinear (aka "defished") output images from the above two shots describe this effect:

Defished and severely cropped Output image A

The center rod seems about straight and slim. So are the rods of the lower (and farther) grid. The two neighboring rods of the top (and near) grid are seemingly bowing (effect of NPP shift) and bulging (effect of perspective and NPP shift, combined) .

Defished and cropped Output image B

Both rods of the top grids appears as if they were bent and bulging while the lower (and farther) grid appears normal and as expected to appear. This difference cannot be amended by application of correction such as PTools coefficients...

NOTE: this "parallax" like error that bowes the shape of the very close straight feature exists even on this single image!

Another real-life illustration of the problem:

This CubicQTVR movie is a composite that shows the strong bowing of the nearest straight object and the way this bowing effect decreases fast when the distance of the next object increases, etc.

The bowing effect is reduced enough when shooting from distance beyond about 60-80 cm to become unnoticeable. This is the reason that allows to make good panography with a fisheye as long as the foremost object is not touching the lens...


Second problem: Blue-Yellow TCA

The fisheye images A and B above were corrected from usual Red-Cyan transverse color aberration (TCA) and the lower grid is therefore very neat after this correction. If you look closely at the rods of the upper grid and especially on the parts against clear white background, you shall see a very strong Blue-Yellow of what seems to be also lateral color.

At the time of this writing I have no explanation to offer for this phenomenon:-(

This "dual TCA" effect on the same image cannot be simply corrected with usual correction tools.


Conclusion

As I could not find any other means to correct the above described two types of defect visible on the output images of claustopanoramas, I had to restore them by hours and hours of editing in Photoshop...

Michel Thoby

7 November 2007