What can 5,000,000 bubbles tell us about face recognition?
Over the past 5 years, my collaborators and
I have studied face recognition using the Bubbles technique (Gosselin & Schyns,
2001; Chauvin, Worsley, Schyns, Arguin & Gosselin, 2005).
I will use this opportunity to take a bird-eye view of this endeavor.
Here is a highlight of what we have discovered: Accuracy and
response time are correlated with the presence of relatively
coarse facial information (mean = 4.5 cpf) between 20 and 110
ms and with the presence of relatively fine facial
information (mean = 8 cpf) between 40 and 110 ms (Bacon,
Gosselin, Vinette & Faubert,
2003; McCabe, Chauvin, Fiset, Arguin & Gosselin, 2005).
But there are important task differences: e.g., discriminating
fearful faces from faces expressing other emotions requires information
at a much finer scale (45-22.5 cpf) (Adolphs, Gosselin, Buchanon,
Tranel, Schyns & Damasio,
2005; Smith, Cottrell, Gosselin & Schyns, 2005). Accuracy
is also correlated with the presence of the right eye between
47 and 94 ms; then with the presence of the both eyes, alternating
with a frequency of about 10 Hz (Vinette, Gosselin & Schyns,
2004; McCabe, Gosselin & Arguin, 2005).
Prosopagnosics (Caldara, Rossion, Mayer, Smith, Gosselin & Schyns,
in press) and patients with damage to their amygdalas (Adolphs,
Gosselin, Buchanon, Tranel, Schyns & Damasio,
2005) do not show this correlation between eyes and accuracy.
Finally, EEG activity at P9 and P10 is mainly correlated with
the contra-lateral eyes of faces
between 150 and 250 ms, irrespective of the
task (Schyns, Jentzsch, Johnson Sweinberger & Gosselin, 2003);
accuracy, response time, and EEG activity at Pz are correlated
with the same visual information (e.g., for happy/neutral face
discrimination: the mouth) between 300 and 400 ms (Smith, Gosselin & Schyns,