Abstract
Using rotating matte and shiny objects, Hartung and Kersten (2002) showed how image motion can affect material appearance. What their demonstrations also revealed, which wasn't however explicitly noted, was that surface material of an object also affects the perceived axis of rotation. For example, a specular teapot appears to rock back and forth while its matte counterpart is perceived as rotating around a vertical axis – though both objects undergo the same rotation. Why is this so? We argue that the perceived axis of rotation of a moving object involves the integration of multiple sources of flow information. Flow from the object contour only (silhouette) can at best provide ambiguous information about an objects rotation axis and at worse give rise to non-rigid percepts. Supplementing contour flow with optic flow arising from the object's material should provide sufficient information to disambiguate the perceived rotation axis, however, flow patterns arising from moving matte textured objects are very different than those arising from specular ones. Here we argue that it is these differences in flow patterns which lead to the differences in perceived rotation axis in the above described phenomenon.
In this work we investigate systematically how 3D shape, contour and surface material contribute to the estimation of the rotation axis and direction of novel, irregular (Experiment I) and rotationally symmetric (Experiment II) objects. We analyze observers' patterns of errors in an orientation estimation task under four different shading conditions: Lambertian, specular, textured and silhouette (Examples: http://bilkent.edu.tr/~katja/orientation.html). Rotation axes were randomly sampled from the unit hemisphere.
Results show, as expected, largest errors for the silhouette condition in both experiments. However, the patterns of errors for the remaining shading conditions differ notably across experiments, yielding larger differences between shaders for the rotationally symmetric objects. We will describe how flow patterns predict these differences.
KD and GK were supported by EC FP7 Marie Curie IRG-239494. RF was supported by DFG FL 624/1-1.