Abstract
Two important monocular cues to depth in static scenes are perspective/object outline and shading. These cues are widely used among artists to indicate depth in flat images, but knowledge of their relative contributions to depth and shape perception, and their interactions, is limited. In order to explore this issue we rendered (using RADIANCE) physically accurate colour images of folded card stimuli and displayed them in 42-bit colour. One side of the card was a saturated red colour, the other white. The sides were separated by a vertical fold to form a concave ‘corner’ or a convex ‘roof’ such that the angle between each side of the card could be varied. A wide range of card angles were produced. Observers viewed stimuli monocularly, through a small circular aperture to exclude other cues to depth, and were asked to match the angle of the folded card stimulus by adjusting the angle between two lines in a ‘view-from-above’ configuration, displayed on another monitor Observers also performed matches to wire frame stimuli (‘outline-cue-only’ condition), large card stimuli that extended beyond the field of view (‘gradient-cue-only’ condition) and stimuli where the angles indicated by the outline differed from that indicated by the gradient (‘cue-conflict-condition’). Results for 3 observers (678 trials each) indicate that the information in the shading is a poor depth cue in isolation, and also ambiguous. Angle estimation seems to be dominated by the perspective cue and a prior for flatness, which can be modelled using Bayesian cue-combination. The addition of shading, of any type, to the ‘outline-cue-only’ improved the accuracy of card angle estimates (linear regression slopes significantly different, p<0.001). Greater improvements in accuracy were observed when the gradient cue was congruent with the outline cue. These results suggest that the gradient cue augments perception of shape from object outline.
Engineering and Physical Sciences Research Council, Grant No. EP/G038597/1.