Detection of binocular disparities

Abstract

A stereo correspondence algorithm designed to perform matching on figurally similar images (arising in normal human binocular vision) is described. It is based on the observation that the operational principles underlying biological stereo disparity detection seem to be extremely general and few in number instead of an extended set of specific “constraints”. We identify one general characteristic of objects in the three dimensional world and use it to formulate a simple noniterative, parallel and local algorithm that successfully detects disparities generated by opaque as well as transparent surfaces.