Abstract
The importance of curvature in visual shape recognition has long been a subject of debate beginning with Attneave's seminal argument that points of maximal curvature are most useful for recognizing shapes (1954). More recently, this result has been both replicated by Norman et al (2001) and challenged by Kennedy and Domander (1985), the latter group claiming that recognition is well served by points of minimal curvature, and still better served by points located between areas of maximal and minimal curvature. The question of curvature in recognition was reevaluated in the current study where subjects were asked to identify a series of shapes displayed as a random presentation of dots placed along what would have been the boundary contour of the shape. Reverse correlation was used to identify the set of points that appeared most frequently during successful trials and also those points that appeared most frequently during unsuccessful trials. In terms of measured curvature, “successful points” identified in this manner did not differ significantly from “unsuccessful points.” In an attempt to confirm the relative importance of each set of points, a second experiment was conducted in which naive subjects who were shown successful points and asked to identify the shape significantly outperformed naive subjects who were shown unsuccessful points. While these data suggest that successful object recognition does not rely exclusively on points of high curvature, it may still be the case that information distributed within regions of high curvature are critical for visual shape processing.