Human face categorization is radially-biased: Fast periodic visual stimulation evidence

At the fovea, face identity recognition preferentially relies on horizontally-oriented face information. Since radially oriented input is privileged in the periphery, along the horizontal meridian (HM) the engagement of face-selective mechanisms is enhanced and subject-level radial bias predicts the speed of saccadic eye movements toward faces. Here, we investigated whether basic-level face categorization is similarly tuned to horizontal information and shaped by the radial bias. Thirty-two participants were tested using Fast Periodic Visual Stimulation (FPVS) combined with EEG. Face images were periodically embedded at 1.2 Hz within rapid streams of non-face objects presented at 6 Hz. This frequency-tagging approach allows face categorization responses to be objectively isolated at predefined frequencies with high signal-to-noise ratio and critically enable dissociation between high-level categorization responses (1.2 Hz, i.e. face oddball frequency) and general visual responses (6 Hz, i.e. base frequency). Stimuli were presented at four peripheral locations (left, right, up, down), while participants performed an orthogonal task to ensure vigilance. Neural response to face oddball frequency were localized on occipitotemporal electrodes with a significant right lateralization. This neural marker of face categorization was strongest along the horizontal meridian. It was also stronger in the upper than in the lower visual field in line with past evidence. In contrast, the general visual response at 6 Hz was maximal along the vertical meridian and in the lower visual field, consistent with known low-level visual anisotropies. Together, these results indicate a clear distinction in visual field anisotropy between high-level face categorization responses, selectively enhanced along the horizontal meridian and in the upper visual field, and low-level visual responses, which preferentially reflect vertical-meridian and lower-field biases. In conclusion, basic-level face categorization is most robust along the horizontal meridian, likely reflecting enhanced access to diagnostic horizontal information. These findings indicate that radial bias selectively modulates horizontal–vertical meridian anisotropies at high-level, category-selective stages of visual processing, following a pattern that is qualitatively distinct from low-level visual responses. Saccades were faster and more accurate towards faces than vehicles. Crucially, the saccadic advantage for faces was largely unaffected by the VM setup, suggesting robust peripheral face detection. However, saccadic behavior was predicted by both the participant's radial bias and their horizontal tuning. This indicates that peripheral radial bias and foveal horizontal tuning jointly contribute to the ultra-fast saccades towards faces.