Abstract
A scheme whereby window operators can be trained to detect the stereoscopic depth of objects in an image is described. Both a theoretical analysis of the behaviour of such windows and some practical results are presented.