Abstract
Perceived shape similarity is central to classic problems in vision science, including object recognition and categorization. Here we consider the "transformational" approach, in which similarity is related to the perceived ease with which one shape can be transformed into another. This definition of similarity can account for asymmetries in which, e.g., shape A is perceived as more similar to shape B than shape B is to shape A. Hahn, Close, & Graf (2009) used morphing sequences of familiar objects to show that when transformations from shape A to shape B are made more salient, shape A appears more similar to shape B, but not the reverse. We demonstrate, using a shape-similarity adjustment procedure, similar asymmetries in shape perception for novel, closed, 2D contour shapes. This result shows that the previously reported asymmetries are at least partly related to generic shape perception. Specifically, they show that neither familiar shapes nor shapes with semantic associations are required to induce perceived shape similarity asymmetries. We further illustrate a general mechanism for inducing perceived shape similarity asymmetry through established perceptual and psychophysical results. In our model, shape similarity asymmetries, like those measured in our experiments, may be explained by biases in visual attention (e.g., due to increasing information at regions of high curvature; Attneave, 1954) and the nonlinear relationship between objective and perceived feature magnitude (e.g., the sublinear scaling of perceived curvature magnitude with objective curvature magnitude; Dobson, 1971). In sum, our results suggest that transformational similarity effects may be, at least in part, a consequence of within-object attention (e.g., Garrigan & Hamilton, 2013) and the perceived magnitudes of salient shape features.
Meeting abstract presented at VSS 2014