Re: Unusual 3d concept

[P3D J.J. Larrea <jjl@xxxxxxxxxxx>]

At 4:16 PM 4/25/96, P3D sam smith wrote:
>I recently aquired a copy of the 1954 publication " Introduction to 3-D", by
>H. Dewhurst, a book mostly  devoted to 3d TV and Motion Picture. While most
>of the concepts I can fathom, there are some that have me perplexed.  On
>page 60, Under the Viewing Aids chapter, it describes a technique similar to
>the parallax stereogram, but using gridded screens.

Sounds like a neat book!  I only know enough about lenticular photography
to be dangerous, so take the following attempt at "reverse engineering"
from such brief descriptions with a grain of salt...

> "  Left and right-eye pictures seen from a fixed stance are photographed
>through a vertical grating in front of the film the width of whose apertures
>was equal that of the bars. Such a film projected on to a screen with a
>similar grating in front would, when viewed at a particular distance, be
>seen stereoscopically."

So they shoot a typical parallax sterogram (pschologram), but instead of
the typical viewing mechanism - put a barrier sheet with the exact same
grid pitch and barrier gap (distance from the film plane) as was used to
shoot the photo, in front of the processed film - they scale everything up
for projection, with a large barrier in front of the screen.  To do this
they would have to scale the grid pitch and barrier-to-screen gap to
reflect the magnification ratio of the projection.

But, since making a parallax stereogram requires knowledge of the viewing
distance (to accommodate for the visual angle subtended between the eyes),
they would have to work backwards through the optical system to determine
the optimal pitch and gap for the taking barrier.  The complete system
includes the enlargement factor from projection to the screen, but also the
reduction factor from screen to audience eyes.

> So far so good. The book then descibes revisions in this process, such as
>"...swinging the camera in an arc centered on the subject with the film
>maintained parallel to the chord of transverse 'pans' the compressed
>slit-type images across the focal planes at the back of each of the
>cylindrical-lens lenticulations of a plastic sheet in contact with the film
>emulsion." OK, I'm a little lost on the swinging camera part.

I think you left out some parts, since now they have switched from using a
barrier grid to using a lenticular sheet when taking the image.  This has
the effect of bending the left- and right-lens images away from each other,
instead of just masking them as in parallax barrier-grams.  It also means
you can get a greater field-of-view onto the same sheet of film, since the
image stripes are laterally compressed.

In a studio situation, the motion of the scene being photographed is under
control, so when shooting stereo there's no need to capture all positions
simultaneously.  Thus instead of 2 different lens positions like a
conventional stereo camera, or 3- or 4- like a modern fixed-lens lenticular
camera, on a lenticular studio-camera a vernier laterally shifts the
lens-board relative to the lenticles-and-film sandwich, and exposures are
made at quite a few relative positions.  Each contributes its own stripe to
the emulsion, and if they are close enough the stripes blend a bit
together.  It is also possible to introduce a corresponding change in the
object with each lensboard position, to get lenticular images which appear
to rotate or animate as you change the viewing angle.

Rotating an object in front of a camera does not look the same as rotating
the camera around the object, due to the background's stillness in the
former case.  To give a sense of "being there" to the audience when they
move their heads back and forth, the entire scene must rotate.  Which means
that the camera must rotate around the focal point of the scene (typically
there will be an object there, since distortions will increase with
distance from the rotational center and you want *something* to appear
undistorted), maintaining a constant distance between the lens's optical
center and that point.  Thus the lens's optical center will describe an
arc.

Something else was left out from the paragraph you quoted, I believe:  The
lensboard must still shift relative to the film sandwich for each exposure;
otherwise all of the captured stripes would hit the same location of the
emulsion, resulting in a blurry non-3d image.  Since it would be nasty to
try to get the arc described above if the lens shifted relative to the
camera body, it is the film carrier which gets shifted between each
exposure to get the relative motion between that and the lens.

What I am not clear on at this point is how you view or project a lenticle-
captured image with a barrier grid, without the image beeing "squished".
Perhaps when calculating the viewing-barrier pitch, one can anamorphically
project, and choose a different barrier pitch relative to the lenticle
pitch than the "1:1 shooting-to-viewing" situation I described earlier
would have predicted, to make up for it?  If they describe any of this in
more detail, I'd love to hear about it.

>Now things get real complicated. It seems the process had a " striation"
>problem when viewing the images.

All these grid pitches and gaps are determined from the viewing distance.
But in a projection situation, that of course varies depending on the seat
you take.  Worse, since the screen is above the head, the viewing distance
will be different at the top of the screen than at the bottom, and even the
gap between the barrier and the screen will change.  This latter probably
introduces nasty moire patterns, as the perceived pitch of the projected
lines will vary relative to the perceived pitch of the barrier grid.

>Somone named Savoye had the following solution:
>" ..Sayoye enclosed his screen by a rotating truncated cone, apex downwards,
>which carried the bars of his grid acrn
>screen. The bars decrease in width towards the bottom, and would pass, if
>produced in imagination, through the same point through which lines in the
>plane of the screen and on the sloping auditorium floor would also
>preferentailly pass. The grid cone rotates at some fifteen to twenty
>revolutions a minute to give an affective occultation frequency of
>forty-eight per second. By such means Savoye has succeeded in eliminating
>the discontinuity in the projected images and made the ' striation' of his
>grid invisible".

How to fix?  Well, choose a target audience member, and determine the
proper optical relationship for the center of the screen, to determine the
pitches and gaps you need.  Now redetermine it for other points on the
screen, for that same viewer:  You will either have to vary the pitch of
the barrier, or the gap between the barrier and the projected image, to
maintain the same optical relationship.  If you assume the barrier is flat,
then the straight rulings would become slightly curvy; but if you want to
use straight rulings, the barrier would have to become a non-planar shape
with continuously-varying gaps between it and the screen.

Now, use the original center point of the screen again, but compute the
proper pitch and gap for that point for each possible seat in the audience.
You'll also get a function that will require either bending the rulings or
shaping the barrier.

Now try to do both: every point on the screen correct for every audience
member.  I wouldn't be surprised if you get a *nasty* function that can't
be solved for all constraints.  Even if simplified it would certainly
require warping in both gap and pitch spaces.  But probably the shape (gap
profile) would roughly involve taking the (flat) barrier, bending it into
an arc to deal with lateral viewing-position changes, and narrowing the
bottom radius to deal with screen-height and floor-slope changes.  Ie.
something like an ice-cream cone in licking position, with the base sliced
off and discarded and then just the front bit of the remainder sliced off
and kept, and the rulings calculated accordingly to compensate for the
narrowing radius.  Which is basically what the paragraph described.

But if the gap-profile determines a shape that can be approximated with a
conic section, with the fulcrum of the cone at the plane of the screen,
then it's certainly possible to not slice off the back of the cone, but
extend it to wrap around the screen.  Of course, since no one can look
through those new sections, it would be silly to do this unless you
intended to spin the cone:

Spinning the cone would cause the image to flicker when the stripes on the
screen went in- and out- of phase with the barrier stripes ("affective
occultation frequency").  Why they want it to flicker, I'm not sure, but
perhaps it utilizes the same human-visual mechanism that comes into play
with film and video projection, where an illusion of continuous motion
occurs when the flicker rate is above a certain number of
frames-per-second?  That's why motion pictures shot at 24 FPS are double-
shuttered in the projector to get 48 FPS flashes, since 48 is typically
considered the bottom limit for motion perception.  So their mentioning
that specific number would seem to confirm this.  Of course, wouldn't the
spinning barrier mean that the image angles are sweeping an angle during
each flash, and wouldn't that blur the picture and screw up the stereo?  I
don't fully comprehend how this works at all well, but if it does it's sure
clever...

Anyway, I'm completely brain-tired thinking about all this, and gotta do
real work now.  Do my rationales make any sense?

- JJ Larrea <jjl@xxxxxxxxxxx>



------------------------------

End of PHOTO-3D Digest 1298
***************************
***************************
 Trouble? Send e-mail to 
3d-moderators@xxxxxxxxxxxxxxx 
***************************