Re: Computer Compositing

[T3D Larry Berlin <lberlin@xxxxxxxxx>]

John Bercovitz wrote:
>I haven't thought out what sort of distortion might be present - I
>just speculated that having an object's associated parallax conflict
>with its depth placement in the scene would produce a distortion
>(amount and type unspecified).  Larry's already noted that he's done
>moderate depth changes and not noticed any distortion so we know (as
>we would expect) that you can get away with it to some degree.

Jim Crowell asks:
Was that with a familiar or simply-shaped object?

>I think the best bet is for me to solve the geometry problem graphically.

It's a knotty question.  Perceived shape-from-stereo depends on vertical as
well as horizontal disparities, and the two seem to be estimated over quite
different areas (vertical are more coarsely measured than horizontal)...

-Jim C.

*******************************************************************

Ultimately a range of objects should be tested with this as well as testing
the resulting images in several viewing options. 

Intuitively the object that is moved deeper into the image will have a
relative object Z dimension greater than would ordinarily be observed. If
the original image was free of distortion, separating it's parts at a
different dimension will not introduce visible distortions except relative
to other parts of an image. In other words, if it's ORTHO in it's original
position, it will still be seen as ortho when displaced farther and made to
appear deeper. As long as these shifts of position leave it within a range
of viewability, the eye/mind still perceives the same object parallax as was
originally recorded. If you shrink the image portion to suggest it's
reduction in size with distance, the relative parallax of the object also
shrinks. In this case you would have to separate the paired pieces for depth
control because shrinking reduces the separation which in (parallel) stereo
makes it move forward instead of deeper.

While perceived shape depends to some degree on vertical changes with
perspective, in a given stereo image, you have a reproduction of
instantaneous perception in which there should be NO vertical disparity.

Larry Berlin


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