Visual Deception Detection using Machine Learning
Keywords
No Thumbnail Available
Authors
Issue Date
2019-01-25
Language
en
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
A recent development in deception detection research is the usage of
computer vision systems. This work analyzes how these systems work
and what the in
uence of di erent input representations is. Three models
are tested, using three distinct abstractions of the same videos from the
MU3D dataset: a model based on the entire image, a face-based model
and a model based on key face regions. Even though the models are
successfully learning, they do not seem to be able to recognize adequate
patterns for a performance signi cantly better than chance. There are no
known prior machine learning applications based on the MU3D dataset.
It is possible that there are not enough deception-related signals in this
dataset. Other conclusions are that the model used is inadequate for this
speci c task, or that all-or-nothing (binary) deception detection itself is
an unfeasible task. An analysis of di erent models and/or on di erent
datasets is required to assert these suspicions.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen