Mailinglist Archives:
Infrared
Panorama
Photo-3D
Tech-3D
Sell-3D
MF3D

Notice
This mailinglist archive is frozen since May 2001, i.e. it will stay online but will not be updated.
<-- Date Index --> <-- Thread Index --> [Author Index]

Re: Computer 3-goD


  • From: P3D Peter Davis <pfd@xxxxxxxxxxx>
  • Subject: Re: Computer 3-goD
  • Date: Thu, 31 Oct 1996 20:38:48 -0500


In order to generate a stereo image pair from a single 2D image in software,
the software would have to:

  1)    Recognize the edges of "objects" in the picture, such as people,
        furniture, etc.  Edge detection is fairly well understood today,
        but it's not perfect.  If an object is in very strong sidelighting,
        for example, the software might pick up the edge of the shadow
        rather than the edge of the object which is in darkness.

  2)    Determine which "shapes" are in front of or behind which other
        shapes.  It might be possible to do this heuristically for a
        specific class of images, where the software could make
        inferences about spatial ordering based on continuity of edges,
        etc.  It would be difficult or impossible to generalize this
        to arbitrary photographs.

        Another approach would be to have the software "recognize"
        objects in the picture.  Again, this could be done for a very
        limited specific class of objects, but today's technology could
        not do this in the general case.

  3)    Generate a displaced view of the scene, offset from the original
        view by, for example, the interocular distance.  This is probably
        the biggest challenge, because it would require constructing
        objects and backgrounds which are not present in the original
        image.  For example, suppose you have a picture of a person
        standing in front of a sign.  In a normal stereo photo, the
        left eye view would contain more of the left side of the
        sign, while the right eye view would contain more of the right
        side.  If you only had one view, how could you possibly reconstruct
        the portion of the sign that was blocked by the person in that
        view?  If the background were simply something textural, like
        grass or distant trees, you might be able to "fake it," but I
        doubt this would work with anything but the very simplest compositions,
        where stereo would hardly make a difference.

-pd
--------
                                Peter Davis
                        http://www.ziplink.net/~pfd/

                   "Nondescript -- the one word oxymoron."


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