Abstract
The classic model of depth processing is of depth tokens added to a 2D map of the spatial layout, forming the “2 1/2 D” sketch of the 3D scene (Marr, 1982). An alternative view is that spatial reconstruction operates by a primordial 3D surface reconstruction that, conversely, embeds the information for spatial layout (Tyler, 1995, Perception). On this latter concept, the multicue surface map contains the only visual localization information (beyond relative phase and orientation cues from local neural filters). These models can be compared by measuring localization ability in stimuli with the depth nulled by opposing disparity against monocular depth cues. If depth is merely an auxiliary variable of the 2 1/2 D sketch, nulling it should not affect localization ability. If the 3D surface map is the inherent substrate of perceptual representation, however, nulling the depth should obliterate the ability to localize targets. Localization of Gaussian profiles sampled at 16′ intervals showed a profound inability to discriminate their location at the depth null point. Moreover, localization based on disparity cues was far better than based on luminance cues alone. These results imply that true localization derives from the depth map rather than in the luminance array, and that Vernier and other high-resolution localization tasks are local luminance discriminations of orientation or phase.
Beyond the primacy of disparity cues, the depth null implies that the contributary depth cues innervate a unitary 3D surface map. Evidence from our own and other fMRI studies suggests that this 3D representation is localized on the lateral surface of the occipital lobe, near the inferior occipital gyrus, originally designated LO by Malach et al. (1995, PNAS) from a 3D-pictures / flat-texture contrast. Activation from the processing of other object properties (structure, symmetry, motion, affordance) occurs separately in more anterior brain areas.