Classifying posed and real smiles from observers' peripheral physiology

Abstract

Smiles are important signals in face-to-face communication that provides impressions / feelings to observers. For example, a speaker can be motivated from audience smiles. People can smile from feeling or by acting or posing the smile. We used observers' physiological signals such as PR (Pupillary Response), BVP (Blood Volume Pulse), and GSR (Galvanic Skin Response) to classify smilers' real (elicited) and posed (asked to act) smiles. Twenty smile videos were collected from benchmark datasets and shown to 24 observers while asking them to make choices, and recording their physiological signals. A leave-one-video-out process was used to measure classification accuracies, and was 93.7% accurate for PR features.