Sunex SuperFisheye TM

What causes this apparent image off-centering on the sensor?

The test image (s)

This a TIFF image converted from a RAW source image shot with an EOS Rebel XT (aka 350D) fitted with a Sunex 5.6 mm SuperFisheye:

In fact you actually see the screen shot of the Realviz Stitcher 5.6 Lens Circle Calibration windows.

No adjustment is yet done here: the red circle A was auto-determined by the software and best estimated by the internal engine.

The red circle A is centered in the image /sensor rectangle as one would expect. The circle is slightly larger than the width of the sensor and this is not expected...

A closer look:

An other circle B with a blue outline can be barely guessed inside the larger A circle.

This blue outline is typical of the peripheral area of a fisheye and I am convinced that it is here also the case: The actual circular fisheye image is restricted into this smaller circle B and one should conclude that it is largely off-centered.

With this assumption in my mind, I have then proceeded into stitching this above hemi-spherical image with its "twin" hemi-spherical image that was shot in the direct opposite direction.

Calibrating the SuperFisheye:

In PTGui Pro 7.3, I have inputed the whole set of 4 images that I have been supplied with by John Law: these four images where shot about 90 degree apart from each other in yaw.

This process allows a good calibration of the lens as the overlap between two consecutive images is wide enough. It would have been impossible to have confidence in the calibration got from only two opposite hemispherical images.

I have cropped the 4 images as described above (circle B) and proceeded with the optimization and PTGui Optimizer (or PTOptimizer) has given the optimized parameters and coefficients for this lens.

I have then subsequently selected a single pair (First and third from left) of opposite images and got this panorama:

Click to see the QTVR (1.4 MB)

The correctness of the parameters and coefficients computed by PTOptimizer seems excellent:

The FOV is estimated to be near to 187 degrees; a= 0 (not optimized) b= - 0.284; c= 0.0666; d= 1.95; d= - 0.08

The questions:

I have checked with PTGui that the circle A would yield a FOV~ 206 degrees (i.e. on par with the Nikkor 10.5 mm). Does this make sens?

Another image was shot later on:

To view the original full size JPEG image please go to:

http://www.sunex.com/dslr/stunning.jpg

Remarks: This image was actually shot about one month after the other four preceding images and with the very same camera. These facts are recorded on the EXIF information of the JPEG and CRW respective photographs.

Observations:

Beware: Due to high compression and heavy sharpening, this JPG image is of a terribly poor quality. It probably doesn't favor the optical ability of the fisheye lens.

At first glance:

With a closer look and after an accentuation of the image Curve left side (PhotoShop Curve) one can see:

Conclusion:

While I can understand that improvement could have been done on the prototype lens to better center the circle inside the sensor rectangle, I see no reason that would increase the diameter of the useful image in such a proportion besides a design change.

The other possibility that I have mentionned (Internal strong flare) cannot be supported with the presently available material... If it was the case, I hope this has been corrected in the final production model.

Not having a sure answer to this question. I wonder if it is wise to review the optical quality of the lens before having images shot with a real production model for test.

Michel Thoby

12 Novembre 2007