Abstract
To interpret the effects of illumination (e.g., shading, shadows, and specularities), observers must estimate the direction, the intensity, and the number of sources in a scene. How this is accomplished is largely unknown, but there is some evidence for a “multi-local” representation that assumes an interaction between local elements (e.g., a moving object and a cast shadow), but no integration into a single coherent lighting model. This allows for multiple point lights that may not be globally consistent, yet taken as a whole, provides an approximation of a global lighting model that reflects the complexity of natural illumination. Such models provide a means for discounting shadows that alter the perceived lightness of surfaces. We investigated this process using lightness judgments in a simple scene composed of a cylinder casting a shadow over a target rectangle centered in a grayscale mondrian pattern. The target patch was entirely in shadow (Exp. 1), or partially in shadow with the remainder covered by a penumbra (Exp. 2) or completely uncovered (Exp. 3).
Observers adjusted the lightness of a reference square (embedded in different mondrian patterns) so that “the target and the reference square were cut from the same piece of paper”. Observers viewed the complete scene, a scene with the cylinder removed (shadow intact), a scene with the cylinder and all information beyond the mondrian pattern removed (shadow cropped), or the cropped pattern from a head-on viewpoint. Observers always perceived the target patch as darker than it actually was in the scene, regardless of the information available about the lighting conditions. However, additional information about the lighting conditions did bring observers closer to constancy, with the largest improvement found for intact shadows (whether the casting object was visible or not). Thus, mechanisms for lightness constancy take into account complex, multi-local scene features in the derivation of a lighting model.