Re: Andrew's paper

[T3D Larry Berlin <lberlin@xxxxxxxxx>]

John Bercovitz writes:
>Larry Berlin wrote:
>> I'd like to point out that Andrew's paper deals with distortions caused by
>> converging cameras and lenses, not the warping of an image after it's been
>> recorded.

>Right, converging cameras was what Michael was talking about.  He said
>his friend thought you could start with a flat photo and take one shot
>of that flat photo by angling in from the left and one shot by angling 
>in from the right and thereby get a stereo pair.
>
>Michael remarked that you would get a warped result due to keystoning.
>That is, you would see a warped flat photo in your stereo viewer.

I thought his friend was mainly theorizing about what could be done in a
computer with flat images to make them 3D. Every simulation of keystoning
(the result of his original thought) that I create in the computer results
in a rotated but still flat plane that represents the original photo.
Perhaps there is another distortion caused by the lens in a camera that
would prevent the result from being flat if you rotate the object and
photograph it. However, since he was a photographer he wouldn't have had to
wait for a computer to try the idea. 

I'm intrigued by the fact that this may be a significant difference between
the computer and cameras. Why would the photos of this rotated image become
warped any differently than the same geometry applied in the computer? Does
the camera create a non-linear shift in the rotated flat image? You almost
have to correct for the keystoning just to get vertical alignment. Would
photos of this rotated picture, corrected for keystoning actually show
visible distortion compared to the same image modified in the same way
strictly by the computer? Or is it simply too small to perceive?

Larry Berlin


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