Processing Facial Expressions of Emotion: Upright vs. Inverted Images

David L. Bimler, Slawomir J. Skwarek, Galina V. Paramei

We studied discrimination of briefly presented upright vs. inverted emotional facial expressions (FEs), hypothesizing that inversion would impair emotion decoding by disrupting holistic FE processing. Stimuli were photographs of seven emotion prototypes, of a male and female poser (Ekman and Friesen, 1976), and eight intermediate morphs in each set. Subjects made speeded Same/Different judgments of emotional content for all upright (U) or inverted (I) pairs of FEs, presented for 500 ms, 100 times each pair. Signal Detection Theory revealed the sensitivity measure d' to be slightly but significantly higher for the upright FEs. In further analysis using multidimensional scaling (MDS), percentages of Same judgments were taken as an index of pairwise perceptual similarity, separately for U and I presentation mode. The outcome was a 4D "emotion expression space," with FEs represented as points and the dimensions identified as Happy-Sad, Surprise/Fear, Disgust, and Anger. The solutions for U and I FEs were compared by means of cophenetic and canonical correlation, Procrustes analysis, and weighted-Euclidean analysis of individual differences. Differences in discrimination produced by inverting FE stimuli were found to be small and manifested as minor changes in the MDS structure or weights of the dimensions. Solutions differed substantially more between the two posers, however. Notably, for stimuli containing elements of Happiness (whether U or I), the MDS structure showed signs of implicit categorization, indicating that mouth curvature - the dominant feature conveying Happiness - is visually salient and receives early processing. The findings suggest that for briefly presented FEs, Same/Different decisions are dominated by low-level visual analysis of abstract patterns of lightness and edge filters, but also reflect emerging featural analysis. These analyses, insensitive to face orientation, enable initial positive/negative Valence categorization of FEs.