Purchase this article with an account.
Kristyn Sanders, Jason Haberman, David Whitney; Mean representation beyond a shadow of a doubt: Summary statistical representation of shadows and lighting direction. Journal of Vision 2008;8(6):334. doi: https://doi.org/10.1167/8.6.334.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Shadows provide information about object motion and depth, but they must also be discounted to accurately recover surface properties and lightness. Our inability to accurately detect contradictory shadows in a scene suggests that shadow information is computed at an early, local scale, and that global factors such as lighting direction are not explicitly represented by the visual system. There is another possibility though. Several recent studies have shown that the visual system represents scenes with summary statistics. For example, observers perceive the average size of a group of randomly sized objects (Ariely, Psych Sci, 2001; Chong & Treisman, Vis Res, 2003), and the average identity of a group of faces (Haberman & Whitney, Curr Bio, 2007). Could shadow information across a scene be represented by summary statistics? To test this, we rendered a set physically realistic images of a simple three-dimensional object illuminated by a single light source from one of 50 different orientations. The resulting images contained a shaded object with a cast shadow (consistent with each of the 50 possible lighting directions). We created a second set of stimuli by converting those same 50 images into two-tone pictures that were not perceived as shadows, but rather as opaque paint. Subjects were presented with a heterogeneous group of either shadow or two-tone images, and asked to compare the mean shadow/lighting orientation to a test image. Mean discrimination thresholds for both the realistic shadows and the two-tone images were precise and surprisingly similar for both stimulus types, demonstrating that, at least for relatively simple scenes, observers can perceive the mean shadow orientation even when there is heterogeneity in the orientation of the shadows. Therefore, although we are insensitive to inconsistencies in shadows and lighting direction, the visual system seems to roughly model global lighting conditions in scenes using summary statistics.
This PDF is available to Subscribers Only