Abstract
The viewpoint invariance of shape recognition is accounted either by proposing three-dimensional (3D) shape descriptions or multiple two-dimensional (2D) representations (obtained from various viewpoints). Almost all of the relevant studies assessing the code for shape representation have based their conclusions on the measurement of rotation effects on absolute reaction times, which can be influenced by low level mechanisms that precede those specifically subserving shape representation. The present experiments used a visual search task to assess directly the code for shape representation by examining whether plane and depth rotations modulate the conjunction (CONJ; decreased search rate when the shape properties of the target are a conjunction of the distrator properties) and the linear non-separability (LNS; decreased search rate when the shape properties of the target are a linear combination of distrator properties) effects, which reflect key features of shape representations (Arguin & Saumier, 2000; Saumier & Arguin, 2003). Experiments 1 and 2 assessed the CONJ and LNS effects when distractor exemplars had either all the same or different orientations, with rotations applied either on the image-plane (Exp. 1) or in depth (Exp. 2). The logic underlying these experiments is that if shape representations are orientation-dependent then the visual properties of the target should no longer be a conjunction or a linear combination of the visual properties of the distractors in the “different-orientation” condition and the CONJ and LNS effects should therefore be eliminated. In contrast, if shape representations are rotation-invariant then the CONJ and LNS effects should be maintained even in the “different-orientation” condition. In both experiments, the results show that the CONJ and LNS effects are unaffected when the distractors are in different orientations. These findings argue for orientation-invariant shape representation codes.