Abstract
Face and object recognition is predicated on the integration of features into wholes. Face recognition is often considered different from recognition of other classes of objects, because integration of face features is “holistic” or simultaneous, whereas integration of other object features is “piecemeal” or sequential. This dissociation suggests differences in spatial and temporal integration between holistic and piecemeal processing, however, face recognition is rarely tested under conditions that demand significant temporal integration. Here, we used a restricted viewing paradigm to force subjects to process face features sequentially. Restricted viewing involved a moving aperture that allowed viewing of only a single feature at a time. In Experiment 1, faces were presented using restricted viewing at five different viewpoints (in-depth rotations) and two different orientations (upright and inverted), and subjects made a two-alternative forced-choice (2AFC) male/female judgment. In Experiment 2, faces were presented at one viewpoint using two different restricted viewing methods and three different whole-viewing methods, and subjects performed a 4AFC identity judgment. In both experiments, accuracy measures showed significant effects of orientation (i.e., “inversion” effects) for both whole viewing and restricted viewing. Viewing-time maps created from subjects' exploration paths during restricted viewing showed that subjects sequentially sampled multiple features, with most of the time spent in areas around the eyes and mouth. The eye/mouth exploration pattern was found even at extreme viewpoints (e.g., 90 degrees) when the nose in profile was very salient and one eye was occluded. There was no difference in the pattern of viewing times across orientations. The results show that the face “inversion effect” exists over a relatively broad time window. The results suggest that “holistic” processing may involve feature integration at multiple time scales.