Many automatic facial expression recognizers nowudoutput individual facial action units (AUs), but several lines ofudevidence suggest that it is the combination of AUs that is psychologicallyudmeaningful: e.g., (a) constraints arising from facialudmorphology, (b) prior published evidence, (c) claims arisingudfrom basic emotion theory. We performed factor analysis on audlarge data set and recovered factors that have been discussed inudthe literature as psychologically meaningful. Further we showudthat some of these factors have external validity in that theyudpredict participant behaviors in an iterated prisoner’s dilemmaudtask and in fact with more precision than the individualudAUs. These results both reinforce the validity of automaticudrecognition (as these factors would be expected from accurateudAU detection) and suggest the benefits of using such factorsudfor understanding these facial expressions as social signals.
展开▼