When otherwise-familiar faces are presented in reversed contrast polarity (e.g. as photographic negatives), they are very difficult to recognize. Here we tested fMRI activity in FFA, in response to quantitatively controlled facial variations in contrast polarity, contrast level, illumination, mean luminance, and rotation in plane. Among these, only reversal of contrast polarity affected FFA activity uniquely. Compared to all other cortical areas, reversal of facial contrast polarity produced the highest fMRI signal change in FFA, across a wide range of contrast levels (5.3 - 100% RMS contrast). By comparison, FFA responses were equivalent (invariant) in response to systematic variations in illumination location, mean luminance, and rotation in plane – even though those parameters also affect facial recognition. In greater detail, reversal of facial contrast polarity changes three image properties in parallel: surface absorbance, shading, and specular reflection. In FFA, we found that the polarity bias was produced only by a combination of all three properties; one or two of these properties in isolation did not produce a significant contrast polarity bias. This suggests that the polarity bias arises from subthreshold (non-linear) summation of multiple face image properties. Using fMRI, we found a homologous effect in visual cortex of awake behaving macaque monkeys. Reversal of facial contrast polarity produced decreased activity, confined to the posterior face patch (homologous to FFA), across contrast levels. Apparently, the polarity bias reflects fundamental mechanisms of visual processing, conserved for at least 25 million years.
This work was supported by the National Institutes of Health (EY017081 to RBHT, MH076054 to DJH), and the National Alliance for Research on Schizophrenia and Depression (NARSAD) (RBHT, DJH).