Abstract
The luminance profiles of opaque and translucent objects exhibit different dependencies on 3D geometry: the shading of Lambertian surfaces exhibits a cosine fall-off as a function of surface orientation relative to the direction of illumination, whereas the light transported through translucent materials declines with the thickness of the object in the direction of illumination. If the visual system relies on these 3D constraints to discriminate opaque and translucent objects, then it should be possible to make a Lambertian object appear translucent (and vice versa) by manipulating its apparent surface orientation and volume. In each experiment, we rendered an image of an object with either Lambertian or translucent material properties, and superimposed sparse random dot fields to generate one of two possible 3D shapes using either motion parallax, binocular disparity, or texture gradients. In Experiment 1, the luminance profile was generated by rendering a Lambertian cone directed at the observer with overhead lighting. Our shape manipulations perceptually transformed the Lambertian cone into a torus, which transformed the appearance into a translucent volume illuminated from within. In Experiment 2, the luminance profile was generated by rendering a back-lit twisted pipe with translucent material properties, and the thickness of the pipe was varied along its length to modulate its brightness. We manipulated the shape to appear as a twisted ribbon, which caused the same gradient to now appear as an opaque matte surface. Although previous work has modulated perceived translucency by manipulating simple image statistics (e.g., luminance contrast and spatial frequency spectra), the experiments reported herein reveal that identical luminance gradients can elicit strikingly different percepts of translucency depending on apparent 3D shape. Together with our previous work on gloss, these results provide strong evidence that the visual system exploits 3D shape constraints to derive material properties from images.
Meeting abstract presented at VSS 2016