Abstract
When the direction of illumination is unknown, the human visual system uses a light-from-above prior to infer the shape of ambiguous shaded images. Can this prior be overridden by contextual lighting cues that indicate the true lighting direction in a specific scene? Previous studies have given mixed results. We re-examined this question using realistic renderings of complex scenes under a range of lighting conditions.
METHODS: Stimuli were computer-rendered scenes of geometric objects. Simulated lighting in each scene consisted of a distant point source and an ambient source. The direction of the point source varied from scene to scene. The strength of the directional lighting cues was varied by modulating the relative strength of the point and ambient sources. Observers judged whether ambiguously shaded disks shown on the geometric objects at various orientations were convex or concave. We used these judgements to infer what lighting direction observers' shape-from-shading mechanisms assumed to be dominant in each scene.
RESULTS: When scene lighting was direct (i.e. point source strong, ambient source weak), the lighting direction assumed by observers closely tracked the true lighting direction. When lighting was diffuse, large individual differences appeared: some observers' shape judgements of ambiguous shaded disks were independent of disk orientation, and some observers had preferred lighting directions that were neither directly above nor in the direction of the direct component of the light source.
CONCLUSIONS: Strong directional lighting cues can override the light-from-above prior: observers interpret ambiguous shaded objects as if illuminated from the direction indicated by the lighting cues. Weak directional lighting can also override the light-from-above prior: surprisingly, the light-from-above prior does not guide shape judgements even in the presence of directional lighting cues that are insufficiently strong to guide observers' estimates of the dominant lighting direction in a scene.