Abstract
Several paradigms have demonstrated a qualitative difference between the processing of faces and other classes of visual stimuli: upright faces are processed as wholes, while other objects and inverted faces are processed as a collection of separable parts. Thus, it has been argued that holistic processing makes faces “special”. The aim of an ongoing project is then to study holistic face processing directly, independent of confounds with part-based contributions to recognition. With standard experimental methods, this isolation is difficult to achieve. However, the existence of a brain-injured patient who retains only holistic processing for faces (CK can recognise upright whole faces, but not inverted faces or objects, Moscovitch et al., 1997) suggests that it should be possible. In McKone, Martini & Nakayama (2001), we introduced a first experimental technique that isolates holistic processing in normal subjects (based on categorical perception of face identity in noise). Here, I describe two very simple techniques that achieve the same aim — one based on identification accuracy with peripheral presentation, and one based on Mooney faces. Each produces good performance for upright whole-face stimuli, together with chance performance for inverted faces and for single face parts; for example, it is possible to find a stimulus position in the periphery where identification of upright faces is still good, but identification of the same faces inverted is completely at chance. To demonstrate the power of these techniques, I describe their use in exploring two properties of holistic face processing, namely its orientation tuning (with image plane rotation) and size tuning. Results have implications for theories of pattern recognition in general, and of face recognition in particular.