Part based representations allow for recognition that is robust in the presence of occlusion, movement, growth, and deletion of portions of an object. We propose a general "form from function" principle arising from the interactions of objects in their environment, which, together with properties of visual projection, gives rise to two kinds of parts: limb-based parts arise from a pair of negative curvature minima with evidence for "good continuation" of boundaries on one side; neck-based parts arise from narrowings in shape. We then test this hypothesis by requiring subjects to partition a variety of biological and nonsense 2D shapes into perceived components. We examine:
1) whether a subject determines components consistently across different trials of the same partitioning task,
2) whether there is evidence for consistency between subjects for the same partitioning task,
and
3) how the perceived parts compare with the parts proposed by the "form from function" principle.
The results are interpreted as suggesting that there are high levels of both intra-subject as well as inter-subject consistency and that a large majority of the perceived parts do in fact correspond to the proposed limbs and necks.
Our proposal for parts envisions a role for them as an intermediate representation between local image features, edges, and global object models; the implications are discussed in relation to figure/ground segmentation, and certain visual illusions.