Abstract
A neural model is presented of how cortical areas V1, V2, and V4 interact to convert a textured 2D image into a representation of 3D shape. Two basic problems must be solved to achieve this end: (1) Transform spatially discrete 2D texture elements into a spatially smooth surface representation of 3D shape. (2) Explain how changes in the statistical properties of texture elements across space induce the perceived 3D shape of this surface representation. The percepts of a fronto-parallel plane, a slanted plane, a cylinder and a sphere, viewed under perspective projection, are simulated for the case of regular-dot surface textures. The sphere example is generalized to the case of prolate ellipsoids where 3D perception as a function of eccentricity is simulated. Results clarify properties of psychophysical data (Todd and Akerstrom. 1987, J. Exp. Psych., 13, 242). In the model multiple spatial-scale filters process the 2D image. Several filters can respond to the same texture features, but to different degrees. The model clarifies how this ambiguous representation of shape is disambiguated using cooperative and competitive boundary interactions that carry out scale-sensitive perceptual grouping within and between filter scales. Across-scale interactions realize a near-to-far depth asymmetry, which has elsewhere been used to explain data about figure-ground separation. These processes take place within multiple, depth-selective boundary webs before the boundary representations regulate the filling-in of a smooth 3D surface representation.
Supported in part by AFOSR, NGA, NSF, and ONR.