UC Berkeley

Face perception is an integral part of social relationships and how humans view the world. Although we typically see faces that range in demographics and expressions under natural conditions, many studies in the field of vision have researched face perception using unnaturalistic, posed, or even artificial stimuli. In this dissertation, I aim to understand how individuals perceive emotional faces as they would in the real world – by using naturalistic stimuli and further examine what features contribute to individuals’ overall perception. Chapter 1 provides a brief overview of the prior literature describing traditional models of face perception in the fields of psychology and neuroscience. These models have historically used controlled, artificial stimuli potentially leading to biases in results specific to stimuli used in those studies. In lieu of using highly controlled face stimuli, I propose using naturalistic images varying in age, gender, expression, etc. to study face perception behaviorally and in the brain through functional magnetic resonance imaging (fMRI). Chapter 2 describes a behavioral study that investigates how individuals judge emotions of naturalistic scenes of crowds. Examining what features in the scene contribute to the overall perception of affect in the crowd, this study revealed that the reliance of the mean emotion of the crowd vs the most emotional face of the crowd varies with the dimension of affect under consideration. Additionally, background context was shown to have an increased influence on reliance of the most emotional face for certain affect dimensions. In the next section, Chapter 3, I investigate how mid-level structural and textural features in naturalistic images of faces are represented in the brain using fMRI. This chapter’s findings show that tuning to mid-level features occurs primarily in retinotopic regions, contradicting traditional hierarchical models of face processing. Furthermore, there exists idiosyncratic differences in mid-level representation – specifically, those with more severe Autistic traits have better representation of mid-level features in retinotopy. In the last experiment, Chapter 4, I seek to understand how high-level semantic features are organized in the brain and whether there exist individual differences in tuning to semantic features. Participants who were worse at face identity recognition were shown to have worse higher identity feature tuning in face processing regions. In the last section of this dissertation, Chapter 5, I summarize my findings from all three experiments and offer insights into these results in the context of the face processing literature.