2D and 3D Components

1
2D and 3D Components
of Face Processing
A mid-level vision analysis
Chen, Kao Tyler (2007, Cerebral Cortex)
2
Faces from the FERET database, filtered in a
4th-power Gaussian aperture and normalized to
the full contrast range
3
Scrambling the phase destroys the local
features global structure but retains the Fourier
energy and net activation of linear receptive
fields
4
Zeroing the phase and reflecting destroys the
local features global structure but retains the
symmetry as well Fourier energy and net linear RF
activation
5
Rotating the face varies the local features and
Fourier energy, but retains individuality and
emotional recognition
6
Inverting the face retains all the local
features and Fourier energy, but reduces
individuality and emotional recognition
7
Symmetry vs. random activation pattern Activation
in mid-lateral occipital regions (replication
of Tyler et al., 2005, NeuroImage)
vs.
8
Face Localizer Faces vs. random activate both
ventral and dorsal regions of lateral occipital
cortex
vs.
9
3D Pose Invariance Ventral face areas are
pose-blind, but dorsal symmetry areas are
activated
vs.
10
Facial Inversion Effect Inverting these full-cue
faces shows a large facial inversion effect in
both ventral and dorsal regions