Action Units of Facial Expressions in Emotional Contagion
Description
This dataset contains processed data from a two-phase experimental study that hypothesized that specific facial Action Units (AUs) could be identified and quantitatively measured to characterize and differentiate facial expressions associated with emotional contagion, specifically laughter, yawning, and mirror pain. The study aimed to determine whether morphometric measurements based on facial landmarks could statistically discriminate between these emotional expressions and neutral states. The research examines facial Action Units (AUs) and their corresponding morphometric measurements through automated facial landmark analysis and electromyographic validation. The dataset includes quantitative measurements derived from facial recordings of participants viewing emotional stimuli without contextual information. Key features comprise normalized Euclidean distances between facial landmarks that characterize specific AUs for each emotional category: AU12 and AU25 for laughter; AU9, AU27, and AU43 for yawning; and AU4, AU7, and AU10 for mirror pain. All measurements were normalized to ensure size independence and comparability across participants. Statistical analyses revealed that the selected facial distances effectively discriminate between the three emotional categories and neutral expressions, with particularly clear distinctions observed in lower face regions. The data corresponds to the analyses reported in the associated manuscript and may be useful for researchers interested in facial expression analysis, emotional contagion studies, or computational approaches to facial expression recognition.
Files
Steps to reproduce
The repository is organized into two main subfolders, each corresponding to one of the two analyses conducted in the study. 1. Expert Approved DS – Facial Pattern Analysis - resources related to the analysis of facial activations based on identified emotional patterns. DATASET DS_JSON_facial_pattern: Contains JSON files with facial landmark coordinates used to analyze facial activation patterns as defined by a FACS coder. SCRIPTS and PROCESSING STEPS - LM_positioning Extracts facial landmark coordinates from each frame . Generates images with overlaid landmarks to visually assess landmark accuracy and JSON files with facial landmarks coordinates for each frames. - distance_matrix_creation Computes Euclidean distances between selected landmark pairs associated with specific Action Units (AUs), as defined by the expert. Produces two matrices: Distances_NO_NORMALIZATION.xlsx → non-normalized distance matrix Distances_min_max_scaling.xlsx → min–max normalized distance matrix -BWV_calculation Computes Between/Within-Class Variability (BWV) for each distance metric. Generates a plot ranking distances in descending order of BWV to highlight the most discriminative features. -Statistics Descriptive statistics: medians and interquartile ranges (IQR) Inferential statistics: Friedman tests, Conover post hoc analyses, and visualizations 2. Raw EMG DS – Emotional Contagion Analysis - scripts used to analyze emotional responses and validate emotional contagion. DATASET LM_coordinates_NEUTRAL → Neutral frames LM_coordinates_STIMULI → Frames showing emotional stimuli LM_coordinates_SUBJ → Participant frames from Phase 2 SCRIPTS and PROCESSING STEPS LM_positioning - Extracts landmark coordinates - Generates visual verification images with landmarks overlaid and JSON files for each frame distance_matrix_creation - Computes Euclidean distances for each frame according to the AUs defined by the FACS coder. - For each dataset (Subjects, Stimuli, Neutral), the script outputs an Excel file containing: Non-normalized distances and Manually assigned frame labels Global Normalization Workflow 1. The script is run separately for Subjects, Stimuli, and Neutral datasets, producing three distance matrices. 2. These three matrices are merged into a single global file: → Global_NO_NORM.xlsx 3. Min–max normalization is applied using global minimum and maximum values. 4. The normalized dataset is then split back into three separate files: - Neutral_min_max_global_scaling.xlsx - Stim_min_max_global_scaling.xlsx - Subj_min_max_global_scaling.xlsx emotional_contagion_validation Violin plots and boxplots Mann–Whitney U tests K-means clustering Technical Information • All scripts were developed and executed using Google Colab → https://colab.research.google.com/ • Python 3.11.11 • All required packages for this environment are included in the folder. • Each script lists the libraries used at the beginning for full transparency and reproducibility.
Institutions
- Universita degli Studi di Torino Dipartimento di psicologiaPiemonte, Torino