Research Article: Using computer-vision and machine learning to automate facial coding of positive and negative affect intensity

Date Published: February 5, 2019

Publisher: Public Library of Science

Author(s): Nathaniel Haines, Matthew W. Southward, Jennifer S. Cheavens, Theodore Beauchaine, Woo-Young Ahn, José A. Hinojosa.

http://doi.org/10.1371/journal.pone.0211735

Abstract

Facial expressions are fundamental to interpersonal communication, including social interaction, and allow people of different ages, cultures, and languages to quickly and reliably convey emotional information. Historically, facial expression research has followed from discrete emotion theories, which posit a limited number of distinct affective states that are represented with specific patterns of facial action. Much less work has focused on dimensional features of emotion, particularly positive and negative affect intensity. This is likely, in part, because achieving inter-rater reliability for facial action and affect intensity ratings is painstaking and labor-intensive. We use computer-vision and machine learning (CVML) to identify patterns of facial actions in 4,648 video recordings of 125 human participants, which show strong correspondences to positive and negative affect intensity ratings obtained from highly trained coders. Our results show that CVML can both (1) determine the importance of different facial actions that human coders use to derive positive and negative affective ratings when combined with interpretable machine learning methods, and (2) efficiently automate positive and negative affect intensity coding on large facial expression databases. Further, we show that CVML can be applied to individual human judges to infer which facial actions they use to generate perceptual emotion ratings from facial expressions.

Partial Text

The ability to effectively communicate emotion is essential for adaptive human function. Of all the ways that we communicate emotion, facial expressions are among the most flexible—their universality allows us to rapidly convey information to people of different ages, cultures, and languages. Further, facial expressions signal complex action tendencies including threat and cooperative intent [1–3]. Unsurprisingly, the ability to produce and recognize facial expressions of emotion is of interest to researchers throughout the social and behavioral sciences.

Our study offers strong evidence that people use discrete AUs to make wholistic judgments regarding positive and negative affect intensity from facial expressions, indicating that patterns of discrete AUs reliably represent dimensions of facial expressions of emotion (analogous to how specific patterns of AUs map to the basic emotions). Our CVML analysis identified AU12, AU6, and AU25 as especially important features for positive affect intensity ratings. Together, these AUs represent the core components of a genuine smile [52]. Note that AU12 and AU6 interact to signify a Duchenne smile, which can indicate genuine happiness [8], and previous research demonstrates that accurate observer-coded enjoyment ratings rely on AU6 [53]. Additionally, the five most important AUs we identified for negative affect intensity map on to those found in negative, discrete emotions such as fear and anger (AUs 4 and 5), disgust (AU9), and sadness (AU4). While AU12 and AU4 have been implicated in positive and negative affect for some time (e.g., [9]), this is the first study of its kind to determine the relative importance of these and other AUs in determining positive and negative affect intensity. Importantly, the strong correspondence that we found between specific sets of AUs and positive and negative valence intensity suggests that contemporary models of constructed emotion may be further tested against basic emotion theories in experimental settings. For example, future studies may investigate the time course of facial expression detection, where basic versus constructed emotion theories make differential predictions on whether basic emotional categories versus emotional dimensions are recognized more accurately and/or rapidly.

 

Source:

http://doi.org/10.1371/journal.pone.0211735

 

Leave a Reply

Your email address will not be published.