One-Eyed Stereo: A General Approach to Modeling 3-D Scene Geometry

Abstract

A single 2-D image is an ambiguous representation of the 3D world- many different scenes could have produced the same image- yet the human visual system is extremely successful at recovering a qualitatively correct depth model from this type of representation. Workers in the field of computational vision have devised many distinct schemes that attempt to duplicate this ability of human vision; these schemes are collectively called "s hape from .... " methods ( e.g., shape from shading, ahape from texture, shape from contour). In thia paper we argue that the di stinct assumptions employed by each of these different achemes must be equivalent to providing a second (virtual) ·image of the original scene, and that all of these different approaches can be translated into a conventional stereo formalism. In particular, we show that it is frequently possible to structure the problern as that of recovering depth from a stereo pair consisting of a conventional perspective image (the original image) and an orthographic image (the virtual image). We provide a new algorithm of the form required to accomplish this type of stereo reconatruc· tion task.