Abstract

How useful are different visual cues in object recognition? An answer to this question depends on the inherent information content of the cue and the effectiveness of humans in using this information. We have developed a method to measure the efficiency (as defined by signal detection theory) with which humans use information conveyed by visual cues in object recognition. The stimuli were four 3-D objects (brick, cone, cylinder and pyramid) seen from any of eight viewpoints. The objects were rendered in orthographic projection in two conditions: (1) as silhouettes (white on a dark background), in which the bounding contour was the only cue to object shape; or (2) with flat shading, which provided surface structure for object recognition. Static Gaussian luminance noise was added to the images, and the signal-to-noise ratio was set so that subject’s performance was not at ceiling or floor. In a trial, the subject was required to name the object. Overall, efficiencies were quite low, less than 1%. Efficiencies were significantly higher for shaded images than for the silhouettes. These two results may imply that shading is used primarily to extract internal contours. This possibility can be tested by measuring efficiency for recognizing line drawings of objects.

© 1991 Optical Society of America