Stereopsis from kinetic and flicker edges.

Abstract

Binocularity seems to appear early in the visual system before any complex form recognition processes take place, The question of the nature of the primitives used for binocular matching is still an issue despite a large number of studies concerned with this problem (e.g., Julesz, 1971; Julesz & Schumer, 1981; Kaufman, 1974). More than a decade ago, Ramachandran, Madhusudhan Rao, and Vidyasagar (1973) argued that a stereoscopic mech~­ ism sensitive to monocularly discriminable contours, Irrespective of the detailed structure of regions bounded by them, is a sufficient-although, as shown by Julesz (1964, 1971), not a necessary-basis for stereopsis. They demonstrated that observers could obtain stereopsis (report the direction of depth of a central patch with respect to i~s surround) by matching of pure texture edges. Stereopsis was also reported under isoluminant conditions using monocularly recognizable stereograms (Comerford, 1974; De Weert, 1979; Ramachandran et al., 1973) and when the matching contours differed in type. For example, Ramachandran et al. found stereopsis in stereograms in which one image contained only texture or color contours and the other only luminance contours. This latter demonstration argues against theories (Frisby & Mayhew, 1977; Mayhew & Frisby, 1976) that try to explain stereopsis in stereograms containing texture contours by a partial point-for-point correspondence in spatial-frequency-tuned luminance-domain channels without any need to involve more complex mechanisms relying on extended monocular processing. In general, results scattered in the literature do seem to suggest that stereopsis can be driven by the results of a rather sophisticated monocular processing based on various edge-detecting mechanisms, that is, primitives mo~e complex than simple zero-crossings or local. extrema In the output of linear filters based on the Laplacian of Gaussian smoothed images (Marr & Poggio, 1979; Mayhew & Frisby, 1981). This conclusion is strengthened by our observation that stereopsis can be achieved in stereograms in which the relevant information is portrayed only by kinetic or flicker edges. Kinetic edges are boundaries defined by motion information alone, without any luminance, texture, or color cues. Our displays were random-dot stereo cinematograms. A set of random-dot stereogram frames was constructed so that two stationary vertical bars were correlated