Abstract
Most ERP studies of object and face perception have focused on the N170, an ERP component that is systematically larger for faces compared to objects in the time window 130–200 ms. We recently demonstrated that N170 effects cannot be explained by differences in amplitude spectrum or stimulus variance, but rather depend on phase information (Rousselet, Husk, Bennett & Sekuler, Journal of Vision 2005, NeuroImage 2007). In the present study, we examined the phase tuning function of EEG single-trials evoked by complex objects. Stimulus phase was systematically manipulated in a parametric design, with 11 steps of phase information, ranging from 0% (noise), to 100% (original stimulus). Contrast and amplitude spectrum were maintained constant across noise levels. Subjects (n=8) had to discriminate between 2 faces, a task orthogonal to the stimulus manipulation. ERPs from each subject were entered into a multiple linear regression model including stimulus phase information, skewness, kurtosis and their interactions as regressors. This simple model explained up to 48% of the variance on average (min=20%, max=67%), with scalp topography very similar to the one of early visual evoked responses. Sharp non-monotonic changes in EEG activity occurred between 100–150 ms. Theses changes were explained by an increased phase sensitivity modulated by the image kurtosis, an interaction that peaked around the latency of the N170. A control experiment using the same task but pink noise (1/f) textures instead of faces did not show EEG phase sensitivity, demonstrating that the effect is not task related. However, wavelet textures preserving higher-order image statistics (skewness and kurtosis) as well as multi-scale phase correlations elicited significant phase sensitivity modulations, albeit overall much weaker and delayed compared to face stimuli. These results suggest that a large part of early responses to complex objects like faces correspond to a rapid bottom-up extraction of higher-order image statistics.
NSERC Discovery Grants 42133 and 105494, the Canada Research Chair program, CIHR fellowship program and the British Academy.