Exploring the veridicality of shape-from-shading for real 3D objects

Abstract

Despite the large number of shape-from-shading studies, the degree of veridicality with which observers perceive real 3D objects from shading cues has not been examined. Six observers viewed semicircular, triangular and trapezoidal corrugations in depth made from gray cardboard of approximately uniform reflectance, presented in a fronto-parallel plane. The object, placed inside a box, was illuminated solely from its top-left or left by a point light source. The corrugations were seen through an aperture that masked their terminating contours and the light source. Observers were asked to draw the depth profile of the object as if it were seen from above. Using a computer mouse, they drew on a computer screen placed just below the object. They also indicated the orientation of the object and the location of the light source. In the first condition, the objects were viewed monocularly; in the second, a white matte sphere was placed in front of the objects to help in locating the light; in the third, the task was repeated with binocular viewing. Drawings revealed that observers were quite accurate in inferring the objects’ shapes when viewing binocularly. There was more variability among the observers when objects were viewed monocularly, with some systematic trends: 1. Most observers were able to recreate the veridical shapes despite monocular viewing. 2. The drawn shapes often differed from the shape of the luminance variations across the objects, thus rejecting heuristics such as “dark is deeper”. 3. In the absence of cues to light direction, observers did not tend to resort to a “light from above” prior. 4. While binocularity heavily aided the extraction of the true shapes, knowledge of the illuminant direction did not. However, even though the exact location of the light source did not aid in extracting 3D shape, when oriented at a horizontal level, the light source did aid in perceiving 3D orientation.