Abstract
Human observers detect faces in the visual environment extremely rapidly and automatically. Yet how basic units of visual information processing, i.e. spatial frequencies (SF), play a role in this remarkable ability remains unexplored. We shed light on this fundamental issue by estimating the minimal and optimal amount of SF content required for fast face detection. Stimulation sequences composed of naturalistic and highly variable images of faces and objects were presented with parametrically increasing SF content (0.50 to 128 cycles-per-image or cpi across 14 SF steps, 4 s/step), such that initially blurry images gradually sharpened over the course of a 56-s sequence. Stimuli were shown rapidly at 12 Hz (83-ms SOA), thereby constraining perception to a single glance. A No Face condition consisted of randomly presented object images, while in the critical Face condition, face images were interleaved among objects every 8th image (OOOOOOOFOOOOOOOFOO…) at a frequency of 1.5 Hz (667-ms SOA). Electroencephalographic (EEG) responses at 1.5 Hz (and harmonics) reflect face detection (i.e. differential perception of faces vs. objects) while responses at 12 Hz (and harmonics) reflect visual processing common to objects and faces (Retter & Rossion, 2016, Neuropsychologia). Participants responded the moment they could perceive faces. All 16 participants detected faces at around 6.46 cpi and showed significant face-selective responses located over (right) occipito-temporal regions in the Face condition only. Critically, this face-selective response emerged at around 4.22 cpi (≈1.69 cycles-per-face or cpf) and steadily increased until 23.24 cpi (≈9.30 cpf). Beyond 23.24 cpi, face-selective responses were equivalent to responses to full-spectrum (unfiltered) faces both in amplitude and spatio-temporal dynamics. In summary, neural face detection emerges with extremely coarse SF information (before explicit behavioural response) but continues to integrate SF content until a relatively fine level of image detail, thereby demonstrating the relevance of higher SF in face detection.
Meeting abstract presented at VSS 2017