The face inversion effect and the anatomical mapping from the visual field to the primary visual cortex

Garrison W. Cottrell, Martha Gahl, Meilu Yuan, Arun Sugumar

Abstract:

The face-inversion effect, or the drastic decrease in accuracy
seen when a participant is asked to identify inverted faces when
compared to upright faces, is an effect that is not found in object
inversion. Here we suggest a new explanation of this effect using
computational models to show that the phenomenon can be
explained by the anatomical mapping from the visual field to
primary visual cortex. We propose that the way inverted faces are
mapped onto the cortex is fundamentally different from the way
upright faces are mapped. Our work first shows the advantages of
this mapping due to its scale and rotation invariance when used as
input to a convolutional neural network. We train the network to
perform recognition tasks and show it exhibits scale and
realistically constrained rotation invariance. We then confirm that
the decline in accuracy seen when a participant is asked to identify
inverted faces is not seen in the network with inverted object
recognition tasks. With the support of these two findings, we test
the face-inversion effect on our network and are able to show the
unique decline in accuracy, suggesting that the way the visual field
is mapped onto the primary visual cortex is a key facet in the
manifestation of this effect.