Abstract
Background: Abstract shape representations must support invariance across a range of 2D transformations as well as certain articulations of the shape. One prominent theory is that shape is encoded as a set of branches along axes of symmetry in the shape (Blum & Nagel, 1978; Feldman & Singh, 2013). Such a representational scheme has numerous attractive qualities, but there is relatively little psychophysical research on its psychological reality. We conducted two experiments seeking evidence for shape skeletons. Design: In Experiment 1, subjects were shown four shapes simultaneously and instructed to select the shape that differed from the other three, with no instruction about how it might differ, receiving feedback after each trial. "Same" shapes had the same number of axial branches, while the different shape had one more or one fewer. We measured subjects' ability to learn to discriminate based on branch numbers. In Experiment 2, subjects performed a same-different task on two shapes briefly presented in sequence. Each "different" trial involved one of three types of changes: metric changes preserving the shape skeleton, a change in the number of skeletal branches, or a change in the number of descendent generations. If human observers employ skeletal representations, we predicted better discrimination performance for skeleton changes than for metric changes. Results: In Experiment 1, performance on the last third of trials did not differ from performance on the first third (p = .192), suggesting that subjects do not learn to attend to branch number differences. However, even in the first third of trials, subjects were better than chance at selecting the different shape. In Experiment 2, subjects were better at detecting skeletal rather than metric changes that led to comparable magnitudes of contour change. Conclusion: These results suggest that skeletal shape representations may exist psychologically but are relatively inaccessible to learning mechanisms.
Meeting abstract presented at VSS 2017