Abstract
An extensive body of literature has aimed to understand how spatial frequencies (SF) and orientations (SO) support efficient face processing (Gold et al., 1999; Duncan et al., 2017). However, few studies have studied both information jointly, and those that did sampled only a fraction of possible combinations of SF and SO (Goffaux et al., 2011). Our study introduces a novel random SF and SO sampling method which applies to the entire spectrum. It is done by applying Gaussian filters centered at randomly selected points in Fourier space, similar to Bubbles (Gosselin & Schyns, 2001), with each point sampling a specific SF and SO. Filters are shaped and distributed in a way reminiscent of the Gabor Rosette Map (e.g. Chen et al., 1996). We used this method in a same/different face matching task, in which 20 caucasian participants completed 2000 trials, evenly split between White and East-Asian faces. A broadband stimulus was presented for 300 ms, followed by a mask. A face filtered using the aforementioned method was then presented for 300 ms, and participants had to determine whether both images were the same. Successful trials were weighted positively, while failed trials were weighted negatively. We then computed a weighted sum of all the presented filters to build a 2D classification image, representing the SFs and SOs correlated with success at the task. A Pixel test using the Stat4CI toolbox (tCrit = 2.3, p < .001; Chauvin et al., 2005) shows that individuals use SF use peaks at 14.5 cycles per face around the horizontal axis (0.8°). This was not influenced by face ethnicity. Our study is the first to examine SO use across face ethnicities, and we aim to replicate this study with Chinese participants to examine possible influences of culture.