Abstract
An intriguing hypothesis in research on shape perception is that shapes are represented in terms of their inferred interior structure, rather than their visible borders. Such 'skeletal' or 'medial axis' shape representations are thought to afford computational efficiency and flexibility. However, an old (and in our view tragically unheralded) report from the 1970s supported the psychological reality of such representations, using a remarkably direct method: subjects were presented with a 2D shape drawn on paper, and they simply used a pencil to make a single dot within the shape. When many subjects' dots were aggregated, the resulting plots bore a striking resemblance to a traditional shape skeleton. Here we revive this paradigm for the digital era, replicating previous effects and extending them in several new ways. Using a tablet computer with a touch-sensitive screen, hundreds of observers were shown geometric shapes (including some that changed size and shape dynamically during viewing) and were simply asked to "touch the screen, inside the shape, anywhere you like". We discovered just when the aggregated touches did and did not reveal shape skeletons — including tests of several types of stimuli that to our knowledge have not been considered in the shape-representation literature. We also employed this method to test predictions made by specific computational accounts. For example, one prominent account based on Bayesian estimation holds that subtle perturbations in a border should not affect the computed shape skeleton. The psychological effects of such perturbations, however, were quantitatively large and visually striking: whereas the aggregated touches in a normal rectangle lined up with the conventional medial axis transformation, the addition of even a very small notch near one corner dramatically altered this pattern. We discuss these and many other results in (re)introducing this surprisingly direct window onto otherwise-hidden visual processes.
Meeting abstract presented at VSS 2012