Politecnico di Torino (logo)

MI-TO: a 3-D database of spontaneous facial expressions

Emanuele Cadeddu

MI-TO: a 3-D database of spontaneous facial expressions.

Rel. Federica Marcolin, Francesca Nonis, Francesco Ferrise. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Biomedica, 2021


The analysis of human emotions has an extremely relevant impact on everyday life. There are many characteristics that make emotions transpire: they can manifest themselves through voice, gestures, or facial expressions. Many researchers have questioned what the most effective methodology is to objectively analyse what the standard markers in the recognition of a particular emotion are and, in this regard, the field of facial expressions has been widely studied. With the advent of Artificial Intelligence (AI), which aims to create a human-computer interaction that improves over time, the study of facial expressions has been directed towards a more automated analysis, which can be used without human intervention. This technology is called Facial Emotion Recognition (FER). FER technology requires the presence of databases aimed at training AI systems. There are many 2-D and non-spontaneous facial expression databases; these, in fact, are composed of images taken from the Web or photographs of posed subjects. However, these are not able to analyse all facets of the face due to the three-dimensionality and spontaneity of expressions. There is a need to fill these gaps, considering characteristics and more realistic aspects of human emotion. The aim of this work is precisely to present a public and embeddable 3-D database, called MI-TO, of facial expressions with the goal of training 3-D FER algorithms. The MI-TO database has been realized in collaboration between the universities Polytechnic of Torino and Polytechnic of Milano. The analyses have been carried out at the 3D Lab of the Polytechnic of Torino. The database (DB) is inclusive of 104 subjects aged between 19 and 35 years, including 49 men and 55 women and mostly of Caucasian origin. The spontaneity of the reactions was researched using 48 images obtained from IAPS and GAPED databases. In order to validate the database, the intervention of a psychologist was also required, who participated both in the choice of the most frames from individual 3D videos and of the use of an algorithm for the recognition of facial expressions. The emotions reflecting the facial expressions in this database are the 6 basic emotions derived from Paul Ekman's studies (anger, disgust, sadness, happiness, surprise, and fear) plus the neutral expression. To achieve the objective just stated, between July 2020 and May 2021, acquisitions of the 104 subjects, now part of the DB, were carried out following a request to complete two distinct questionnaires aimed at measuring the levels of alexithymia and empathy, to assess their suitability for the study. Then, the participants were subjected to a video analysis, while they were shown in random order the 48 images extracted from the two validated databases. For every image, each participant was also asked to respond to the SAM scale to give a judgment on the subject shown in terms of valence and arousal. Once the acquisitions were completed, each video was analysed in order to extrapolate RGB and Depth images, which were then inserted into the database. The comprehensive database features 104 subjects, who differ in the variability of facial expressions, and more than 4.500 downloadable images. The Mi-TO database is an important breakthrough in the research of facial expression recognition in terms of geometric analysis and diversification of spontaneous emotions, showing strong interest in future work thanks to its integrability, favoured by the simplicity of technology acquisition and low material cost.

Relators: Federica Marcolin, Francesca Nonis, Francesco Ferrise
Academic year: 2021/22
Publication type: Electronic
Number of Pages: 112
Additional Information: Tesi secretata. Fulltext non presente
Corso di laurea: Corso di laurea magistrale in Ingegneria Biomedica
Classe di laurea: New organization > Master science > LM-21 - BIOMEDICAL ENGINEERING
Aziende collaboratrici: UNSPECIFIED
URI: http://webthesis.biblio.polito.it/id/eprint/21720
Modify record (reserved for operators) Modify record (reserved for operators)