Was "seeing the mean emotion" indeed a high level analysis?

Abstract

Introduction Previous research found that observers could precisely extract summary statistics from a broad range of visual stimuli, including facial emotion. However, it is controversial whether the results found in the mean emotion discrimination task indeed showed a kind of high level analysis. The task was adapted from the low level ensemble coding, like mean size (Ariely, 2001; Chong, & Treisman, 2003). Does it really reflect the high level extraction of mean emotion? Methods Two experiments were designed to address this question. Fifty morphed faces were created from two emotionally extreme faces (neutral to disgust) of the same person. In Experiment 1, the same task of Haberman and Whitney (2007, 2009) was used. Observers were asked to judge whether a single test face was more neutral than the preceding set of four faces with identical or different emotions displayed for 2000ms. In Experiment 2, observers made similar comparison, but based on the size of eyes within the faces. Results The threshold analysis used by Haberman and Whitney (2007, 2009) didn’t show a consistent pattern for every subject. Therefore, A’, the nonparametric sensibility index of signal detection theory, was calculated as the index of discrimination performance. The results showed that although observers could indeed precisely discriminate the mean emotion of four hetero faces from the emotion of the test face, discriminating the emotion of homo faces was much better. Besides, compared with previous results (Haberman, & Whitney, 2007, 2009), the mean discrimination performance showed no differences among upright, inverted and scrambled faces (Haberman, & Whitney, 2007, 2009). More importantly, the mean extraction of eye size was as precise as that of emotions. Conclusion It was possible that the summary representation of facial expressions might be based on low level, instead of high level, visual analysis of faces. Meeting abstract presented at VSS 2013