Abstract
When segmenting an object into its perceptual parts, certain parts appear more ‘distinctive’ than others. The trunk of an elephant, for example, is more distinctive than its tail, partly because few animals possess a trunk, whereas many have tails. These distinctive parts can be a vital clue for category membership. But even in novel objects, in which such real-world knowledge can not be applied, limbs vary in their distinctiveness, indicating that shape processing plays an important role in identifying the likely diagnostic features of objects. Which factors determine what makes a part distinctive? To investigate this, we showed 9 observers over 500 2D silhouettes composed of a large main body and two parts, differing in size, number of concavities, curvature characteristics and other features. They were then asked to indicate which of the two parts was the more ‘distinctive’. Overall agreement between observers was well above chance at 83%, showing that observers consistently used similar strategies. They generally preferred parts with a larger area, more concavities, and where the contour curvature statistics deviated more from the main body. These results give us first insights into how we evaluate and make sense of novel objects and might explain how humans can generalize classes from a small number of samples.