Abstract
Studies of shape-from-shading have shown that humans tend to assume that stimuli are lit from above. Such studies typically use ambiguous stimuli; the stimulus orientation for which perception is bi-stable indicates the preferred lighting direction. We take an alternative approach using classification images as a tool to probe the lighting prior. This technique uses either target stimuli embedded in a strong noise mask or noise only stimuli and asks observers to detect some target feature. Noise samples are accumulated according to the observers' responses and the resulting images represent the observer's template for the target in question. We asked observers to identify the image containing a small bump (they were told that the diameter would be around 5 cm / 2.5 deg) in a 2-IFC task in which only 1/f noise images were shown (no target was presented). Observers were not show example targets but rather were asked to imagine the image that would be formed by frontal viewing of a shape presented to them, once only, as a profile sketch. Observers' templates were derived by averaging the noise samples leading to positive responses and subtracting the average of those samples that were not selected. There were two control conditions large bumps (notional diameter 8 cm) and white disk. Templates were typified by approximately Gaussian white blobs. We measured the position of the luminance peak in each template. Peaks in ‘bump’ templates were typically offset from the centre compared to ‘white disk’ templates (mean offset for ‘small bump’ = 0.25 deg). ‘Large bump’ templates had larger offsets (∼0.5 deg). There was a tendency luminance peaks to be above centre in ‘bump’ templates. This result is consistent with a template based on a shaded bump lit from above.
Supported by EPSRC Grant EP/F026269/1.