Abstract
The visual system efficiently extracts summary statistical information from visual scenes (e.g., sets of oriented gabors, crowds of faces). Ensemble representation precision may in part be driven by the power of statistical averaging, which can be leveraged to overcome noise at the individual item level (Alvarez, 2011). The current experiments explicitly tested this by comparing the representation of noisy individual faces with that of noisy crowds. If the ensemble computation inherited noise from the individual item level in a linearly additive process, we would expect ensemble representations to suffer more with the introduction of noise. However, this was not the case. Observers viewed either individual faces or sets of 4 faces masked in varying degrees of noise (no noise, low, medium, high) and had to adjust a fully visible test face to match the perceived individual or average expression. The difference in precision between the no noise condition and the noise conditions was analyzed in a 3 (noise) x 2 (set size) repeated measures ANOVA. This revealed a main effect of noise, but no effect of set size, indicating that performance did not depend on whether observers assessed a single face or a crowd of faces. The interaction was also not significant. Overall, the results are consistent with the notion that ensemble representations may be more robust to noise than might be expected based on the representation of individual items, a useful outcome given the often noisy inputs available for the ensemble calculus.
Meeting abstract presented at VSS 2018