Abstract
Contours can be generated by any one of several possible physical events, including texture, occlusion and specular reflectance. Previous research has shown the classification of contours as texture depends on the pattern of shading adjacent to contour boundaries (e.g., Marlow, Kim & Anderson, 2011). These textural boundaries preserve shading gradients across edges generated by local changes in albedo (Ben-Shahar & Zucker, 2001). Unlike textures, occlusion boundaries have very different constraints that cause shading discontinuities (Ben-Shahar, Huggins & Zucker, 2002), including extremal shading toward the contour that is informative for making figure-ground distinctions (Palmer & Ghose, 2008). It is possible that the differentiation of occlusion boundaries from textural boundaries might depend generically on variation in shading direction across contours. We undertook a series of experiments to systematically vary the shading across reflectance boundaries generated by the surfaces of 3D rendered objects. We initially painted our surface with a dark and light texture similar to military camouflage. The lighting direction was parametrically varied to generate changes in shading across reflectance boundaries. For example, we rendered the lighter regions with a light source directed from above (at 0 degrees) and the darker regions with a light source directed from different orientations ranging up to 180 degrees. We refer to this change in the direction of local gradients as the delta shading angle. Increasing the size of delta shading caused the edge to no longer appear as a reflectance boundary, but instead, appear as an occlusion boundary. We find that the side of the contour perceived as figure or ground depends on a light-source from above prior. The resulting apparent break in surface continuity is also sufficient to support amodal completion. We conclude that the classification of contours, and the interpretation of gradients that define them, depends on more than just low-level visual processing.
Meeting abstract presented at VSS 2016