Abstract
The ability to extract summary statistics from a set of similar items, a phenomenon known as ensemble perception, is an active area of research. In exploring high-level ensemble domains, such as the perception of average expression, researchers have often utilized gradually changing face morphs that span a circular distribution (e.g., happy to sad to angry to happy). However, in their current implementation, face morphs may unintentionally introduce noise into the ensemble measurement, leading to an underestimation of ensemble perception abilities. Specifically, some facial expressions on the morph wheel appear perceptually similar even though they are positioned relatively far apart. For instance, in a morph wheel of happy-sad-angry-happy expressions, an expression between happy and sad may not be perceptually distinguishable from an expression between sad and angry. Without accounting for this perceptual confusability, observer error will be overestimated. The current experiment accounts for this by determining the perceptual confusability of a previously implemented morph wheel. In a 2-alternative-forced-choice task, 7 observers were asked to discriminate between multiple anchor images (36 in total) and all 360 facial expressions on the morph wheel (which yielded close to 27,000 trials per participant). Results are visualized on a 'confusability matrix' depicting the images most likely to be confused for one another. This confusability matrix reveals discrimination thresholds of relatively adjacent expressions and, more importantly, uncovers confusable images between distant expressions on the morph wheel, previously unaccounted for. By accounting for these 'confusability regions,' we demonstrate a significant improvement in model estimation of previously published ensemble performance, suggesting high-level ensemble abilities may be better than previously thought.
Meeting abstract presented at VSS 2017