Abstract
Critical spatial frequencies (SF) that enable visual object identification change with stimulus size (Majaj et al., 2002; Oruc & Barton, 2010). Observers utilize coarse features at small sizes and fine details at large sizes. Faces viewed at small sizes around 2° width are recognized using SFs around 3-4 cycles/face-width—a scale too coarse to resolve individual facial features. In contrast, faces larger than 5° in width are recognized using SFs around 8 cycles/face-width. However, it is unclear whether these results, which are due to subthreshold viewing of noisy images, are relevant to suprathreshold recognition of familiar faces. We employed a celebrity-naming procedure to examine this question. Observers (N=19) were asked to name 100 celebrity faces in a random order. Each face was viewed in one of four conditions based on size (small=2° vs. large=10°) and resolution (intact vs. blurry). Blurry faces were generated by low-pass filtering the images at 6 cycles/face-width. Faces were randomly assigned to each condition for each individual observer and viewed in a random order. Faces that were unfamiliar to the observers (determined in post-test debriefing) were excluded from the analysis. A repeated-measures ANOVA showed a significant main effect of size (F(1, 18)=17.11, p< 0.001) and resolution (F(1,18)=44.17, p< 0.001) as well as a significant interaction between the two (F(1,18)=17.36, p< 0.001). Intact faces were named with near 100% accuracy at both sizes. Accuracy for blurry faces was 84% at the small size, which was significantly greater than the 73% accuracy at the large size. Blurry face-naming accuracy was greater at the small size compared to large for 15 of the 19 observers. These results have implications for approaches to visual aids for low-vision. Specifically, they suggest that magnification may not be an appropriate option for improving face recognition.
Meeting abstract presented at VSS 2018