Abstract
Evidence from schematic stimuli (Rensink and Cavanagh, 2004 Perception) suggests that the detection of discrepant shadows in a visual search task is impaired in upright, but not inverted, images - hypothetically due to the discounting of shadows. The current study investigates this phenomenon with real objects and their shadows.
Sixty pebbles were photographed from directly above as they were rotated through four orientations with direct (ringflash) and indirect illumination that cast shadows. The visual difference between all pebbles (without shadows) was estimated using a visual difference predicting model (Párraga, et al 2005 Vis.Res. 45, 3145; Lovell et al 2006 ACMTAP, 3, 155). Classical multi-dimensional scaling was applied to the resultant confusion matrix and the two major axes, corresponding to size/color and orientation, were identified.
Visual search stimuli featuring 5, 10 or 15 pebbles were constructed. Distractor pebbles all cast a shadow in the same direction, while the target pebble's shadow was rotated by 180°. The heterogeneity (based upon the MDS axes) of the pebbles within each stimulus was also varied. Stimuli were viewed monocularly, and were presented in a fronto-parallel arrangement. However, texture gradients were added to induce a perception that the pebbles were upon a ground, ceiling or fronto-parallel surface. Stimuli were presented upright (light from above) or upside down (light from below). The observer's task was to locate the discrepant shadow.
There was a significant effect of image inversion: visual search was significantly slower in the light-from-below condition. There was also a significant effect of heterogeneity: search became slower when the heterogeneity of the pebbles was increased. Texture-gradient had little effect. The apparent discrepancy between the results of Rensink and Cavanagh (with small, 30° changes in shadows) and ours (with 180° changes) may illuminate characteristics of the mechanism that underlies shadow discounting.
Supported by EPSRC/Dstl (GR/S56405/01 & GR/S56399/01).