Exploring Contagious Emotions via Ekman’s Action Units

Emotional contagion reveals the fascinating way emotions spread among individuals. Through verbal and nonverbal signals, one person's emotions can influence the emotional atmosphere of others, creating a shared ambiance and enhancing empathy and interpersonal connection. In recent years, interest in this process has significantly increased due to scientific research, and an increasingly important role has been attributed to the synergy between psychology, physiology, and the application of artificial intelligence in these fields. One of these innovative directions involves the application of machine learning techniques in the field of facial expression recognition. In this context, advanced computational algorithms can categorize emotions represented in photographs with increasing accuracy. This achievement is made possible through access to image databases containing facial expressions linked to various emotions. However, a large portion of these databases focuses on a limited number of basic emotions, including joy, sadness, anger, disgust, fear, and surprise. Moreover, attention is often directed towards images, with few video resources available, despite their extreme usefulness in emotional contagion research. In conclusion, these databases often present artificial and inauthentic representations of various emotions, a significant limitation for studies aiming to provide authentic and spontaneous stimuli. From the collaboration between the Politecnico di Torino and the psychology department of the Università di Torino, the idea of creating a new database emerged. This database aims to provide authentic video stimuli of six emotions or actions, directly linked to rarely represented emotional states in currently available databases. The emotions involved include laughter, mirror pain, yawning, embarrassment, contempt, and schadenfreude. The main goal is to offer the scientific community an original database with carefully validated video stimuli, capable of determining their effectiveness as stimuli for future studies on emotional contagion. It is during the validation phase that the significant importance of a multidisciplinary approach became evident. On one hand, expertise from the psychological field played an essential role in addressing the complex landscape of emotions. In parallel, engineering methodologies were employed to integrate artificial intelligence to develop a series of classification tools for emotions, based on both images and electromyographic signals collected during experiments. In the following thesis, I present one of the three engineering approaches that have emerged as a result of our collaboration. Specifically, my objective is to create a tool with the capability to identify emotional contagion by analysing facial images, using a classifier that can differentiate between the various emotions we considered in our study. To achieve this, I conducted a detailed examination of the typical facial movements associated with each emotion, with a particular focus on the concept of action units, muscular movements that play a defining role in expressing an emotion. This investigation ultimately led me to devise a classifier based on geometric distances between specific characteristic points on the face, called landmarks, automatically positioned through the application of artificial intelligence. The method used demonstrated remarkable accuracy in classifying the considered emotions and in detecting the activation of the associated action units.