Abstract
When presented with unfamiliar faces that vary in expressions, angles, and image quality, observers exhibit high rates of recognition errors (Jenkins et al., 2011). Specifically, in unconstrained identity-sorting tasks, observers generally struggle to "tell people together" (cope with variation across different images of the same person) while being able to successfully "tell people apart" (distinguish between images depicting two different people; Andrews et al., 2015). The use of ambient, noisy face images in this simple card sorting task both reveals the magnitude of these face recognition errors, and suggests a useful platform to re-examine the nature of face processing using naturalistic stimuli. Currently, we used this task to assess the impact of two stimulus properties (image blur and orientation) known to affect face recognition in tasks using controlled stimuli, but which may have a different impact when applied to ambient face images. In Experiment 1 (Image Blur), we recruited 64 participants who sorted highly-variable images of unfamiliar female faces subject to low (N=22), medium (N=21), or high levels of blur (N=21). In Experiment 2 (Picture-plane Inversion), we recruited 35 participants who sorted either upright (N=17) or inverted (N=18) images. We analyzed the sorting solutions from both tasks using signal-detection descriptors (Balas & Pearson, 2017) that allow us to characterize card grouping in terms of sensitivity to extra-personal variability and response bias. We found that in Experiment 1, the level of image blur did not modulate performance, affecting neither observers' d' values (F(2,61)=1.51, p=0.23) nor their response criterion (F(2,61)=1.82, p=0.17). In Experiment 2, however, we found that picture-plane inversion led to significantly poorer sorting performance than the upright condition (t(33)=2.91, p=0.0065) but no change in the response criterion (t(33)=-1.52, p=0.14). Unconstrained, ambient face identity sorting is thus resilient even to substantial levels of image blur, but suffers greatly from face inversion.
Meeting abstract presented at VSS 2018