Abstract
When we view a 3D scene, object features are seen on 3D surfaces infused with lightness and color at the correct depths. Most models of 3D vision do not explain how this happens. A 3D LAMINART model proposes how laminar ortical mechanisms interact to create such 3D surface percepts using interactions between boundary and surface representations (Grossberg and Howe, 2003; Grossberg and Swaminathan, 2004). The present work develops the model to predict how textured images with multiple potential false binocular matches, such as dense stereograms, generate the correct 3D surface representations of igures and their backgrounds. In addition, the model shows how, when textured stereograms define emergent occluding and occluded surfaces, the partially occluded textured surfaces may be amodally completed behind the occluding textured surface. The model hereby provides a unified explanation of data about stereopsis and data about 3D figure-ground separation and completion of partially occluded object surfaces. The model clarifies how interactions between layers 4, 3B, and 2/3A in V1 and V2 contribute to stereopsis, and proposes how binocular and monocular information combine in V2 and V4 to form 3D boundary and surface representations. The model extension includes an important role for surface-to-boundary feedback to eliminate many spurious boundaries that could otherwise interfere with correct object recognition at inferotemporal cortical processing stages. In particular, a disparity filter and 3D cooperative-competitive grouping laws are predicted to interact with 3D surface filling-in operations via feedback between V2 pale and thin stripes, respectively, to overcome this problem. This interaction also triggers figure-ground separation of occluded and occluding objects, and helps to convert the complementary rules for boundary and surface formation (Grossberg, 1994) into a consistent visual percept.
AFOSR, DARPA, NSF, and ONR