Surface but not volumetric part structure mediates three-dimensional shape representation: evidence from part-whole priming.

The decomposition of three-dimensional (3-D) objects into shape primitives consisting of geometric volumes is a key proposal of some theories of object recognition. It implicitly assumes that recognition involves volumetric completion--the derivation of a three-dimensional structure that comprises inferred shape properties, such as surfaces, that are not directly visible due to self-occlusion. The goal of this study was to test this claim. In Experiment 1 participants memorized novel objects and then discriminated these from previously unseen objects. Targets were preceded by primes containing a subset of object surfaces that either matched those visible in the whole objects or that could only be inferred through volumetric completion. The results showed performance benefits through priming from visible surfaces but not from inferred surfaces. In Experiment 2, we found equivalent priming for part-primes containing two visible surfaces from the same volumetric part and for primes containing one surface from each of two volumes. These results challenge the view that 3-D object recognition is mediated by shape primitives comprising geometric volumes. Instead, the results support an alternative model that proposes that 3-D shapes are represented as a non-volumetric surface-based structural description.