[photo-3d] Quiz no. 2 - Stereoscopic Deviations

["Dr. George A. Themelis" <drt-3d@xxxxxxx>]

This is perhaps the trickiest to identify.  At first 
sight it appears that a mounting error is involved.  
If you place a horizontal ruler over the pair you will 
notice that the background is out of alignment (left 
picture is higher than right).  However, the foreground 
is in good alignment.  So, how can that happen?

What makes this pair especially difficult to diagnose is 
the lack of middle ground.  If there were objects over 
the entire depth range, you would notice that the 
misalignment is zero at the foreground (window level) 
and gets worse as you move away, being maximum at infinity.  
This observation suggests that the stereoscopic deviations 
are vertical instead of horizontal.

So, we flip each member by 90 degrees, and voila!  
A perfect stereo pair!

Now, "perfect" is in the eyes of beholder.  Some people would
say that the "correct" pair (recorded with the camera rotated 
by 90 degrees) is pretty BAD!!!   However, in my book, turning 
the camera by 90 degrees results in a technically sound stereo 
pair because it fits the definition of a stereogram (small 
deviations in the horizontal direction).  Whether you like the 
result or not is a different issue.  It looks unnatural because 
we are not accustomed to walk around with our heads tilted 90 
degrees. 

One moral of the story:  If you attempt to correct a slightly 
titled picture by counter tilting the chips, you are introducing 
a vertical error.  You can think of the stereoscopic deviations 
as vectors which, in a properly recorded and mounted stereo
pair, lie in the horizontal direction.  Stereo projection
or anaglyph pairs give you a nice opportunity to study the
orientation of these vetors.

Tilting the pair during mounting makes this vector have a 
component in the wrong (vertical) direction.  The error is 
worse the larger the vector (more depth) or the larger the 
tilt.  But, from a practical point of view, for small tilts 
and moderate depth ranges this error is very small and 
sometimes better than a tilted picture.

George Themelis