Abstract
Li and Zaidi (2003) showed that when 3-D developable surfaces are covered by isotropic random-dot textures, fronto-parallel concave surfaces are seen as convex, because orientation flows are absent, and the spatial-frequency gradient is consistent with a convex percept, if image spatial frequency is assumed to vary solely as a function of distance. The percept suggests that deformations of texture elements were not used by the visual system. Haptic feedback can influence priors and weighting of 3-D visual cues (Ernst etal, 2003; Adams etal, 2004)). We tested whether haptic learning can correct the perception of 3-D shape based on spatial frequency cues. Observers were shown four half-cycles of sinusoidal corrugations (Convex, Concave, Right-slant, Left-slant) and a flat surface, all covered with a random-dot pattern. Observers perceived concavities and convexities as convex, both slants as concave, and the flat surface as flat. Using a Phantom force-feedback device, observers were then allowed to “feel” the actual 3-D shapes. After repeated exploration, observers started perceiving the concave and slanted surfaces “correctly”. The effect of haptic learning spread over the complete image, but disappeared when the haptic feedback ended. Since the texture in each image is physically compatible with the veridical shape, this may be due to recruiting correct cues or to overriding texture cues. Since the frequency cues are similar for the two curvatures and the two slants, haptic feedback indicating the opposite curvature or slant predictably evoked the percept compatible with haptic information. As a critical test we presented flat-fronto parallel haptic feedback for the textured images of the curved and slanted surfaces. This feedback failed to modify the pre-training percept. In addition, curved or slanted haptic feedback did not alter the percept of the flat stimulus. Consequently, prolonged haptic training can disambiguate texture cues to 3-D shape, but can not override them.