Abstract
How does the visual cortex process complex natural scenes? The 3D LAMINART model has been developed to explain many data from psychophysical experiments in terms of how laminar cortical mechanisms interact to create 3D boundary and surface representations (e.g., Cao & Grossberg, 2005, Spatial Vision; Fang & Grossberg, 2007, Spatial Vision; Grossberg & Yazdanbaksh, 2005, Vision Research). Here the model is extended to show how the same mechanisms, properly refined, can explain how the brain generates 3D surface representations in response to complex natural scenes. The model describes how monocular and binocular cortical cells interact with later stages of 3D boundary formation and surface filling-in in the LGN and cortical areas V1, V2, and V4. It proposes how interactions between layers 4, 3B, and 2/3 in V1 and V2 contribute to stereopsis, and how binocular and monocular information combine to form 3D boundary and surface representations. The major challenge for processing natural scenes is that the 3D boundaries are often unconnected and incomplete, and that cluttered surface regions incorporate many possibilities of false binocular matches. The main new developments are: (1) feedback interactions between V1 binocular cells and V1 surface cells help with initial depth assignments, notably how V1 surface filling-in helps to enhance V1 monocular and binocular boundaries; (2) feedback between V2 boundaries and surfaces completes broken boundaries and eliminates false binocular matches; and (3) a V2 disparity filter in both the boundary and surface processing streams together help to generate correct 3D surface representations. The model hereby provides a unified approach to providing both a quantitative explanation of data about 3D stereopsis and surface perception of psychophysical displays, as well as a system for 3D processing of natural scenes in computer vision applications.
Supported in part by the National Science Foundation (SBE-0354378).