On Thu, 17 Sep 1998 02:37:40 +0200, Helmut Dersch wrote:
>Neuman-Ruether wrote:

>> (you even seem to agree with me now [which you did not
>> appear to do with your first post...;-] that we do see in
>> curves, and further, that "small objects look more natural in
>> the fisheye image..." [I would have said "rounded objects"
>> rather than small - though, I guess, as I think about it,
>> even straight-sided objects look OK if small in a fisheye image,
>> so maybe your choice of words is the better one...;-], that we
>> don't come to the same (rather obviously necessary, as it
>> seems to me, given what goes before the conclusion...;-)
>> conclusion about the preferable imaging-perspective type
>> for representing what we see in a wide field of view on a
>> small imaging surface...! ;-)

>Well, I have to admit, that I did change and refine some
>of my points, but not the key issue. Take a look at the
>portrait images on
>
>When I generated them I was very surprised myself how well
>the fisheye-portrait matches the "true" 80mm-lens portrait.
>I knew it would beat the rectilinear image, but I thought
>the panoramic image would be better, which it clearly is
>not.

I was surprised by this too... I have thought of the
swing-lens panarama camera as a good compromise, and since one
is essentially presented with what amounts to just the center
strip of the lens projection everywhere in the image (the narrow
center strip of the lens projection is all that is recorded), it
would have seemed that this would have look most like the 80mm
view...

>Now comes the point (again and again :-) : This match has absolutely
>nothing to do with our peculiar vision system! Any animal, any robot,
>any image processing software, any reasonable human or non-human
>would agree that image 3 (the fisheye portrait) matches the real
>view (image 1) best!

Uh-huh...! ;-)

>> In the full-frame 16mm view on your web page at:
>> http://www.fh-furtwangen.de/~dersch/perspective/Wide_Angle_Perspective.html
>> if people would imagine the image with approximately the top
>> third and bottom 1/4 removed, the sides rounded, the edges softened
>> and blended to black, and the area away from the extreme center
>> softened in focus, it may look rather familiar as an image type...;-)
>> Restoring the image to the form in which it appears on your web
>> page is not an unreasonable thing to do for the sake of producing a
>> small image for viewing, I think, and does not destroy its essential
>> value as a way of representing the way we see...

>I agree on this: We could take advantage of the errors of our vision system
>(ie seeing bent lines where there are straight ones in reality) to fake
>our brain in thinking a small image represents a wide angle view;
>haven't tried that yet. It would be similar to the use of ultrawide angle
>lenses in architecture which fake our brains in thinking rooms are tall
>when they are small (not due to errors in our vision, but due to
>the wrong viewing distance). A pure fisheye projection probably doesn't work
>because large objects are just too much distorted. I will try a couple
>of "intermediate" transformations with my software.

Like the super-wide I suggested that has 1/2 the fisheye curvature,
huh? ;-) I think this would be nice... BTW, there is an advantage to
the curvature in our vision (and in using fisheyes, or at least very
wide lenses with strong barrel distortion, for my video work), so for
me, the curvature is not an error in vision, but quite a valuable
characteristic.... (though of course it is an error if one is trying
to establish whether a line "in reality" is straight - but looking
straight at it solves that problem). The curvature reduces the extreme
magnification changes near the edges as the viewpoint is rotated, and,
if building verticals are viewed at normal distances (and the
buildings are not too high), the viewpoint can be tilted with less
change in the convergence of parallel lines - in other words, rotating
the viewpoint in any direction can be done more smoothly, without
resulting in a lot of the weird size-change effects that wide-angle
three-point rectangular perspective would show in rotation (which
would be as unsettling while seeing "reality" as it is when seeing
video or film shot with a super-wide "well-corrected" lens that is
being tilted or panned...;-) Also BTW (though I may have mentioned
this earlier), the Greek architects were well aware of the eye
curvature of off-axis straight lines, and tried to deceive the
viewers of temples into thinking that they were either larger than
they were, or that the viewers were closer than they were, by
curving the upper horizontal lines a bit and introducing some
curvature and convergence to "parallel" vertical lines
(neat trick! ;-).

>As to your particular proposal: Do you mean like
>?
>Interesting, but would also be OK with a panoramic
>(swing-lens type) camera. The real challenge is to create a natural
>looking image with large field of view in either direction.

Interesting image, but I guess I overstated the cropping...;-)
Maybe 1/5th off the top, nothing off the bottom, with the
soft edge somewhat more curved at the sides...(?) Ugly image,
but an interesting experiment! ;-) Thanks for the try!
I would like to see a 15-16mm lens for 35mm with strong barrel
distortion (but well short of full-out spherical perspective).
Or maybe an 18mm (?) fisheye orthographic projection lens, but
with the far edges (where much of the curvature happens) cropped
out...