Re: Computer Compositing

[T3D Bill Stratemeyer <wwstrat@xxxxxxxx>]

john bercovitz wrote:
> > An object's
> > own associated or shall we say "internal" parallax is a function
> > of its distance from the observer so I think if you change the
> > distance of the object, some type of distortion must result.

Bill Stratemeyer wrote:
> >  I just finished a test of modifying deviation only for
> > objects in a stereo pair.
> [...]
> >  When viewing and comparing the 2 sets of stereo pairs the
> > only difference appears to be the depth of the dolphin with no
> > distortion (Bulge or stretch) and the object size seems to be
> > the same as in the unmodified pair.
> [...]

john bercovitz wrote:
> Bill, I would expect distortion in the theoretical sense.
> Whether or not that distortion is observable would depend
> on a number of factors including the amount of distortion
> there and also observer-related variables such as object
> familiarity.

  I agree as the near object (dolphin) Has a lot of "internal"
 parallax I expected there to be a lot of distortion.  But given
 that it is a long tubular shaped soft edged object it probably
 is not the best test subject. Even Zooming in to 2X and viewing 
 only the dolphin did not reveal any obvious distortion,
 particularly in size as the top and left side of the images 
 have a scale to compare object size.(the scales should be homologues)

> Can you describe your process in a little more detail?  That
> way maybe we could try to judge the amount of distortion
> present.  When you say 0" or .25" of deviation, what's that
> referenced to?  That is, how much deviation do you have for
> infinity homologues and how much for the stereo window?

  The images I used were from a set done for Lenticular
printing.  This required the camera to be converged to a point
mid depth in the scene.  This is to limit total on print 
deviation in the foreground and yield reverse deviation 
that is not pseudo in the background.(lenticular
double depth?)
  
 The deviation figures I gave are the difference in the near 
objects position compared to the convergence point. 
(homologues that appear at the window, surface of the monitor
or print)
 They also have the same deviation difference when measured from 
the left side of the image....now I see a window problem here, 
will window disparity cause distortion in an object if homologues 
are properly spaced for cross-eyed viewing. 
  
  So the control images are flawed with some distortion but it
is too small to see, Key stoning should be the most offensive.

  This calls for a proper test with objects of various shapes
and sizes with parallel translation of the camera.  Any suggestions?
I can render a scene with just background to use for quick
cut & paste of missing background information so several
test pairs can be readily generated. 

Bill Strat...


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