Abstract
Purpose: How well does the visual system maintain constant perceptions of object reflectance properties across illuminant changes with complex spatial structure? Methods: Three observers (one naïve, two experienced) adjusted the diffuse (body color) and specular (glossiness) components of one grayscale sphere to match the appearance of another. Spheres were rendered using grayscale versions of the Debevec (SIGGRAPH98) light fields and presented on a high-dynamic-range display. For symmetric matches, both spheres were rendered using the same light field, but randomized mirror reversal of the light field was introduced to minimize reliance on local image features. For asymmetric matches, a different light field was chosen for each sphere. We studied conditions where the spheres were viewed in isolation and where each sphere was viewed within a scene rendered with its corresponding light field. The task type (symmetric or asymmetric) and context type (no context or complex context) were presented in blocked order. In addition, we varied the diffuse and specular components of the standard sphere. Two settings were collected for each task-context-standard combination. The accuracy of subjects‘ settings was quantified in terms of an error magnitude. This was defined as the magnitude of the deviation of a setting expressed as a percentage of its veridical value. Error magnitudes were computed separately for the diffuse and specular components of each match and averaged over observers and choice of light fields. Results: We have collected preliminary data for three observers and three light field changes. Symmetric matches were close to veridical, both when the spheres were viewed in isolation (6% error diffuse; 13% specular) and when they were viewed within a scene context (6% error diffuse; 10% specular). Asymmetric matches were worse when the spheres were viewed in isolation (24% error diffuse; 68% specular). When each sphere was viewed in a scene context, there was a slight improvement in asymmetric matching performance, relative to isolated viewing (16% error diffuse; 65% specular). This latter effect varied with observer and light field change. Conclusions: The spatial structure of the illumination affects the perceived lightness and glossiness of 3D objects. This effect appears to be reduced by contextual information that provides cues to the illuminant.