Re: That Truck Again

[P3D Larry Berlin <lberlin@xxxxxxxxx>]

>Date: Tue, 09 Sep 1997
>From: P3D Jonathan Gross continues:
>.....................
>What you say is true for monoscopic images.  The problem is that with
>stereoscopic images pairs, the subtle high frequency variations in color
>between the left and right fields may be all that defines horizontal
>boundary differences (parallax) for a particular feature. .............
>.............. Assuming regression towards the dominant color in
>the area in question, we now might loose as much as 3.1% (4/256*2) of
>the detail which would have created stereoscopic parallax.


If parallactic information is very small and only in the vertical dimension,
your argument might have greater validity. As it is, most stereo parallax
information is diagonal features, lines and curves that extend far more than
2 or 3 pixels. What is missing in one very small area due to the process you
describe, is made up by adjacent areas where the dominant color *does
belong* to essential parallax information/edge identification. When viewed
in stereo, the eyes and mind are able to connect essential data despite a
level of loss. The lower the compression, the less the loss. It takes quite
a bit of compression for this effect to even be noticed. The result is that
the loss in stereo images is still larger than in monoscopic images but not
nearly as great a problem as you seem to make it out to be.

Larry Berlin

Email: lberlin@xxxxxxxxx
http://www.sonic.net/~lberlin/
http://3dzine.simplenet.com/


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