Abstract
To recognize a shape, the representation of a currently available shape must be matched to the representation of a previously presented shape. Many studies have investigated characteristics of the stored representation, but relatively few have looked at what types of shape are more or less difficult to represent. In a series of experiments, we measured how quickly subjects could learn to recognize various types of shape (closed contours, open contours, and segmented contours) and tried to infer the shape characteristics that predict better learning. Over 9 alternating training and test sessions, subjects attempted to learn to recognize 16 novel shapes. During each training session, subjects viewed each of the 16 shapes twice while making same/different shape judgments. During each testing session, subjects viewed the 16 shapes from the learning set and 16 new shapes, all presented in a random sequence. Subjects were asked to label each shape as either a member of the learning set or a new, never before seen shape. We found that open 2D contours were just as quickly learned as closed 2D contours, suggesting that shape learning is insensitive to the global and structural differences between these shape types. In contrast, we found that separating the contours into two parts significantly impaired learning, suggesting that shape learning is sensitive to contour connectedness.