Abstract
Our interaction with the physical world requires the visual system to estimate and represent the 3D geometry of objects from various visual cues. Image motion is a particularly powerful cue to 3D structure, as has been investigated in many experiments using classical shape from motion (SFM) stimuli. However, these studies commonly assume - implicitly or explicitly - a diffusely reflecting surface. While optic flow in these cases is directly linked to the object motion; its behavior when the surface is specular, the specular flow, is related to the 3D curvature of the underlying shape (Koenderink & van Doorn, 1980). While this specular flow may facilitate a complete reconstruction of the 3D shape in the theoretical sense (Adato et al. 2010, 2011), its different nature may also bias the perceptual estimation of 3D shape from motion. Indeed, in previous work we have shown that specular flow systematically biases the perceived objects' rotation axis (Dörschner et al 2013), an implicit computational step in SFM. Here we investigate whether the perceived 3D shape of moving specular objects differs systematically from that of matte-textured ones. Stimuli were bumpy, 3D wavers rendered either with a diffusely reflecting texture map or as a specular surface. During presentation these objects rotated back and forth through 20 degrees, or they were kept static as a control condition. This design resulted in 2 (motion, static) x 2 (matte, specular) trial types where observers (N=7) were asked to judge local shape and curvature at indicated locations by adjusting a rotation-invariant shape index probe. Results indicate that shape estimates differ between static and moving conditions, across different reflectance properties, and across locations on the object. We account for these results with a computational approach to specular shape from motion.
Meeting abstract presented at VSS 2014