Abstract
We present the shifty shade illusion, a new motion illusion with important implications for computing correspondences in apparent motion displays. Participants reported 2AFCs or ratings of perceived motion after a sequence of two frames. Frame 1 (F1) contained a rectangle divided into 12 vertical bars, consisting of 4 repetitions of a luminance triplet of light (L), medium (M) and dark (D) bars (e.g., LMDLMDLMDLMD) against a homogeneous background of one of 12 luminance levels (black to white). All six possible luminance orderings within the triplets were tested. In Frame 2 (F2), the center bar within each triplet was deleted, revealing the background. When L-bars were deleted in F2 against dark backgrounds (>.6 Michelson contrast from white), participants perceived motion toward the M-bars in the F1-F2 transition, whereas they perceived no motion when the L-bars were deleted against light backgrounds (<.6 contrast). Deleting the D-bars had the opposite effect, producing motion toward the M-bars against light backgrounds and no motion against dark backgrounds. Deleting the M-bars always caused motion percepts, behaving like L-bars on dark backgrounds and D-bars on light backgrounds. The illusion persists when there is no change in contrast polarity across edges from F1 to F2, suggesting it is not a case of reversed phi (Anstis, 1970). The results are better predicted by the change in contrast asymmetry between the center region's left and right edges from F1 to F2 (r=.92) than by Adelson and Bergen's (1985) motion energy model (r=.74). For example, rightward motion was perceived when the degree to which the center bar contrasted more with the right bar than the left bar was greater in F2 than in F1. This edge-based pattern persisted even when surface-based luminance correspondence predicted otherwise. The results suggest that edge-based correspondences are more influential than surface-based correspondences in determining motion perception.
Meeting abstract presented at VSS 2014