Generalization of the Features of Emotional Faces

Abstract

Emotional faces contain important social information and draw attention automatically. Some emotional expressions draw attention more rapidly than others; however, various studies disagree as to which emotional expression consistently draws attention the fastest, with both positive (e.g. happy) and negative (e.g. angry) emotional expressions showing faster reaction times depending on the study and task. This brings into question whether differences in reaction time to emotional faces are due to valence alone or factors such as low level image features (e.g. contrast and orientation). Additionally, if these low level features, particularly spatial frequencies, are involved in the rapid processing of emotional faces, then non-face objects with similar spatial frequency content would have similar reaction time effects. In this study, we examined the role of spatial frequency content in access to awareness of images of emotional faces. We used car images to test for generalizability based on low level features. Using the spatial frequency content from angry, happy, and neutral faces, we used machine learning to classify car images, both frontal and side views, as happy or angry based on their spatial frequency content. Using breaking continuous flash suppression (b-CFS) and a forced choice task, we measured reaction time to access of awareness as well as participants' subjective rating of images of emotional faces and classified "emotional" car sides and fronts. No significant differences were found between either image type or emotion in b-CFS, and notably, faces did not reach access to awareness faster than car images. In the rating task, however, human faces were rated as expected (e.g. happy faces as happy) even though car images were rated neutrally.