Abstract
Surface material constancy under variations in object shape, illumination, and viewpoint is a remarkable achievement of the human visual system. Here, we investigated how the assumed viewing distance to a surface influences material recognition. A set of photographs depicting surfaces with ambiguous material identity were shown to observers, who provided estimates of apparent viewing distance and judgements of surface material for each image. In subsequent experiments, separate groups of observers were given contextual information about the spatial scale of the photographed scenes, either with explicit instructions (e.g., “the camera is very far from the surface”) or with objects of familiar size that were digitally inserted into the images. Our experiments demonstrate that for a subset of these images, the extent of between-group disagreement about the material identity of a surface is greater when observers have conflicting assumptions about the spatial scale of the scene. These findings also indicate that materials can have canonical spatial scales; that is, the plausibility of recognizing a given material category is constrained by its characteristic appearance across changes in viewing distance. Significantly, heuristics that exploit the interdependence of material appearance and viewing distance point to a previously unacknowledged generative constraint on the statistics of images that are diagnostic of material classes.