Abstract
An open issue in visual object recognition concerns the structure and organisation of long-term memory (LTM) representations of object shape. While much research effort has focussed on the so-called ‘iewpoint-debate’ relatively less work has examined the structure of shape representations. Here we test contrasting hypotheses about the degree of abstractness in shape representation -examining, in particular, the extent to which intrinsic shadow caused by scene illumination affects the recognition process. One hypothesis is that LTM shape representations are shadow invariant, and that filtering shadow from the image during perception incurs a time cost. Alternatively it may be that while LTM representations are shadow invariant, shadow provides information that facilitates the recovery of 3D object shape. An alternative hypothesis is that LTM representations are shadow-specific and that changes in shadow will incur costs resulting from mismatches in the memory representations. These contrasting claims were examined in two experiments. In Exp 1, Ss memorised four targets rendered with intrinsic shadow from a larger set of novel objects. Recognition memory was examined in a target/non-target discrimination task using primes containing no shadow or intrinsic shadow. The results showed that priming was shadow-specific, being larger for primes containing the same intrinsic shadow as the memorised targets (for both familiar and novel viewpoints). In Exp 2, an object matching paradigm was used in which Ss matched novel objects for shape across variations in shadow. RTs were faster for objects with no shadow than for those with intrinsic shadow. This pattern of results suggests that (1) at some level, LTM representations of objects are shadow-specific consistent with some recent hierarchical image-based models (e.g. Riesenhuber & Poggio, 2002; Current Opinion in Neurobiology, 12, 162–168), (2) intrinsic shadow information can also hinder he recovery of 3D object shape during perception, presumably because it increases task complexity.