Visual Perception of Material Properties

Abstract

Visual experience allows us to make instant, context-dependent judgments about material properties of surfaces, aiding in recognizing and categorizing materials based on their functional utility. This thesis explores gloss perception and material recognition through investigations into how visual sensitivity to surface properties like specular reflectance varies under different conditions, and how our ability to discriminate and categorize materials is influenced by bottom-up and top-down visual processes. The first study examines the scaling and discriminability of perceived gloss, focusing on how gloss sensitivity changes with the magnitude of specular reflectance and the role of internal sensory noise. We find that a model based on Maximum Likelihood Difference Scaling (MLDS), a suprathreshold perceptual scaling method, can efficiently and accurately estimate just-noticeable differences in specular reflectance, indicating that human gloss sensitivity operates with the same additive internal noise assumed by the model. The second study proposes a novel framework for measuring gloss sensitivity, assessing how observer rankings of gloss differences can inform the development of a measurement standard for gloss appearance in diverse contexts. We find that observers consistently rank gloss differences resulting from variations in lighting, shape and viewpoint, while the physical reflectance (in this case, roughness) of the surface is held constant. An image-computable model of visible differences (HDR-VDP-3) accurately predicts observer rankings of suprathreshold gloss differences, which can establish reasonable bounds on gloss sensitivity across these viewing conditions. The third study demonstrates that, for certain images in which the material identity of surfaces is ambiguous (e.g., a field of wheat vs. a patch of carpet fabric), observers with different assumptions about apparent distance assign different material categories to describe the surfaces. This material-scale ambiguity, a previously undocumented phenomenon in visual perception, has implications for theories of material recognition and categorical perception in general. Together, these studies suggest that the visual perception of material properties relies on estimating image features diagnostic of material category under typical viewing conditions. Under ambiguous viewing conditions, however, these image features are more open to interpretation than previously thought.