Cornell University
SMILES AS SIGNATURES: INVESTIGATING IDENTITY IN FACIAL EXPRESSIONS
Abstract
dc:description.abstractFacial expressions, particularly smiles, have long been studied for their emotional and communicative functions. This study investigates whether smiles also carry identity-specific information, functioning as signatures unique to individuals. Using a dataset of over 19,000 images of 38 celebrities, we extracted Action Units (AUs) using Openface (Facial Action Coding System) and employed both logistic regression and neural network models to classify individuals based solely on their smiles. Results demonstrated that smiles enable individual identification at a significantly higher accuracy than chance, 18.66% with logistic regression and 21.8% with a neural network, compared to a chance level of 2.63%. In contrast, classification based on neutral expressions yielded lower accuracy (4–9%), further supporting the idea that smiling expressions encode identity-specific cues. Feature elimination and AU analysis revealed six key AUs (4, 6, 7, 10, 12, and 14) that were most predictive of identity. These findings suggest that smiling serves not only as a form of expression but also acts as a part of individual identity. This research contributes to emerging perspectives in expression-based biometrics, proposing that how we smile may say as much about who we are as what we feel.
Degree
thesis:*- Name thesis:degree_name
- M.A., Psychological Sciences and Human Development
- Level thesis:degree_level
- Master of Arts
- Discipline thesis:degree_discipline
- Psychological Sciences and Human Development
- Grantor
- Cornell University
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Lamba, Manmeet Kaur
- Chair dc:contributor.chair
-
- Anderson, Adam
- Committee member dc:contributor.committeemember
-
- DeRosa, Eve
Subjects
dc:subject × 6Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Bibid
- 17422623
- Dc Identifier Other
-
ProQuest Submission ID: 12545
ProQuest Publication ID: 32170333