Re: Stereo Window and Convergence Examples

[P3D John W Roberts <roberts@xxxxxxxxxxxxxxxxx>]

>Date: Tue, 24 Jun 1997 00:08:11 -0500
>From: P3D Larry Berlin  <lberlin@xxxxxxxxx>
>Subject: Re: Stereo Window and Convergence Examples

>**** Well described. It made me realize a part of why I consider convergence
>to be a powerful cue. While you are viewing a stereo image, you can direct
>your full attention to very small areas of the total scene. Doing so
>provides a measure of the convergence of any specific point (eye muscle
>comparisons). Checking both a close and distant point would define a
>triangulated sense of space for the whole scene. 

>...I seldom look at a stereo image without testing the convergence of many
>points in the scene. It's a quick and almost automatic scanning process.
>That's why I consider it to be a strong effect. I use it all the time. I
>find that doing so sensitizes one to smaller variations and this enhances
>the overall depth effect by correlating with associated retinal disparities.
>The two types of sensation reinforce each other.

I think Larry's description is the closest to my views on the subject.
Some folks have been claiming that convergence is a "weak" depth cue and
retinal disparity is a "strong" depth cue - how can that be when when it is
impossible to stereoscopically judge depth without a *combination* of the
two? In the words of Richard L. Gregory (Eye and Brain - the Psychology
of Seeing, chapter 5):
   "There is a clear linkage between two of the depth signalling mechanisms
   we have described - (1) the convergence of the eyes serving as a *range
   finder*, and (2) the difference between the two images giving *disparity*.
   The angle of convergence adjusts the scale of the disparity system. When
   the eyes fixate a *distant* object, disparities between the images are
   seen as representing *greater differences in depth* than when the eyes
   are converged for near vision.
 
   "If this did not occur, distant objects would look closer together in
   depth than near objects of the same depth separation, for the disparity
   is greater the nearer the objects."

I think part of the problem is that there has been no effort here to 
rigorously define "depth perception". The "convergence is weak" crowd seem 
to be implicitly assuming that "depth perception is nothing more or less 
than the ability to sense that one object is nearer or farther than another
object". The author cited above appears to take the view that depth perception
is the ability to judge the distance to an object. I would tend to agree more
with the author of "Eye and Brain" - I think of depth in terms of distance
rather than order of proximity.

It should be noted that the brain does not use feet (or meters) as a natural
unit of measure, so estimates made in those terms will have limited accuracy.
Also the limited "resolution" of convergence sensing affects all distance
estimates. I think Gregg Podnar and George Themelis have made good points
on the relative nature of depth perception, especially when viewing a stereo
photo. When I view a stereo photo, I usually "shut off" the feedback cue
pertaining to absolute convergence, and I don't think of what I see in the
scene as being a particular physical distance from me. I generally think of
the depth of the image in terms of itself, or perhaps relative to the stereo
window when that's a convenient reference point (mostly just relative to
the image - the window would be secondary). However as Larry described, I use
both convergence and retinal disparity to judge the depth in the stereo photo.

There is evidence that other people view stereo photos in this way. For
example, there have been many posts recently on the convention of mounting
a stereo pair so that the nearest object in the field of view is at the
stereo window. Depending on how far the subject was from the camera when the
photo was taken, this may or may not be "realistic" - for instance the
"infinity" of the photo may be much "closer" than "real infinity". If
people actually insisted on the greatest possible realism, this would not
be tolerated - the stereo window would have to be set to keep image infinity
close to true infinity. But the observer still uses *relative* convergence -
the convergence cue is not shut off entirely.

I don't know why this switch to relative convergence is so easy - perhaps
the decoupling between convergence and accommodation is a cue that this isn't
a "real" scene. 

Note that infinity doesn't actually have to be in the scene for there to be
a distortion in nearby objects (based on absolute convergence).

John R


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