Abstract
Last year, Rivest and Moscovitch (2002) introduced a prosopagnosic man (DC) who, despite his impaired face recognition, has intact object and word recognition. His face processing was here further evaluated using a face matching task developed by Wilson, Loffler and Wilkinson (Vision Res. 42, 2909–2923, 2002). Photographs of 19 different faces served as target faces, and each was geometrically transformed into a synthetic face. At each trial, a target face was presented with 4 synthetic choice faces, and DC and DCB (his brother) had to select which choice face matched the target one. The target and choice faces were presented both in front views, both at 20 side views, and the target and choices, at 20 side and in front views, respectively. Like neurologically intact observers (previously tested by Wilson et al.) and DCB (93.8% correct), DC obtained 97.5% correct when matching front view faces. Similar to DCB (87.5%), he obtained 83.8% when matching side view faces. However, on side-front matching, DC obtained 57.5% correct, a performance worse than that of DCB (81.3%), and other controls (90.7% in Wilson et al). We conclude that DC's intact part-based processing system is sufficient for matching same view faces but not for matching faces oriented 20 apart. This deficit suggests that the face-specific holistic system is necessary to help a robust representation of faces across rotations in depth. The results confirm that the synthetic faces developed by Wilson et al. provide sufficient geometric information to make accurate discrimination of faces.