Abstract
So far, the contribution of spatial frequency (SF) and orientation to face perception has been explored separately although these dimensions are jointly encoded during primary visual processing. Namely, it has been shown that SF around 12 cycles per face (cpf) and horizontal face structure optimally convey face identity, respectively. In the present experiments, we delineated the primary visual information most useful for face perception by manipulating the content of face images both in orientation and SF. Picture-plane inversion was used to evaluate the face specificity of the observed SF-orientation effects. We asked participants to match upright and inverted faces that were filtered in the Fourier domain. Faces in a pair contained horizontal, vertical or the combination of horizontal and vertical ranges of information in low (between 2 and 8 cpf), middle (between 8 and 32 cpf) or high SF (between 32 and 128 cpf). The advantage for processing horizontal compared to vertical bands of face information arose in SF located between 8 and 16 cpf. It decreased at lower (2–8 cpf) and higher SF (32–128 cpf) ranges. Picture-plane inversion decreased the horizontal advantage mostly between 8 and 16 cpf. Our findings suggest that the expert processing of upright faces is selectively tuned to horizontally-oriented face structure carried by intermediate SF bands. This SF-orientation tuning is lost with inversion indicating that it results from observer-dependent face-specific mechanisms.