Lei Yuan¹ (), Claudia Wu¹, Ipek Oruc¹; ¹University of British Columbia

Previous work has shown that faces with a high degree of blur are more recognizable at smaller sizes. In addition, blurry faces generate identity adaptation aftereffects at small, but not large, sizes. This small-size advantage for blurry faces has been observed using digitally scaled and blurred face images. Here, we examine whether the small-size advantage persists in a paradigm with greater ecological validity, where size variation is achieved through viewing distance, and blurring is achieved via physical blurring filters. We tested 12 participants (5 females, 7 males, ages: 20-44 years, M=26.67, SD=7.16) in a face recognition protocol at two viewing distances (1m and 0.35m) corresponding to small (1.66 degree) and large (4.75 degree) face sizes. At each trial, a celebrity face was displayed on a computer screen until response. Trials were blocked by viewing distance (close vs. far), where a randomly selected half of 100 celebrity images were viewed at each distance. The order of blocks was counterbalanced across participants. Faces were viewed behind a 1-degree Luminit holographic Light Shaping Diffusers (LSD®), such that blur level was fixed throughout the experiment. Participants were asked to identify the celebrity at each trial. If the participant was unable to correctly identify the face, they were immediately shown intact images of the celebrity to determine whether they were unfamiliar with the celebrity, in which case the trial was discarded. Face recognition accuracy was 66.22% at the small size, significantly higher than the 60.19% at the large size (p=0.014, d=0.39). Ten out of 12 participants showed the small-size advantage at the individual level. These results replicate earlier findings of better recognition for blurry faces at smaller sizes with digital manipulation of blur and size, and extend them by demonstrating that this small-size advantage generalizes to physically blurred faces viewed at varying distances.

Acknowledgements: This work was supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) USRA award (LY), an NSERC Discovery Grant RGPIN-2019-05554 (IO) and an Accelerator Supplement RGPAS-2019-00026 (IO).