Computer Science Department
School of Computer Science, Carnegie Mellon University


Spontaneous Human Emotion Recognition from Video

Mounira Tlili

M.S. Thesis

August 2019


Keywords: Deep Learning, Machine Learning, Emotion Recognition, Emotions, Facial Expressions, Micro Expressions, Facial Action Coding System, Action Units, Lie Detection, Psychology, MobileNet, Depthwise Convolutions, 3D Convolutions

From the computer vision perspective, the problem of automated facial expression analysis is a cornerstone towards high level human computer interaction. In this research we are interested in detecting true (non acted) human emotions for applications such as automatic lie detection, psychological diagnostics and the detection of malicious intents in public spaces. We use a guided emotion detection and identification approach based on facial muscle contractions and relaxations. Our method is composed of two main parts: (1) Action Unit 1 detection which reaches a very high average precision per Action Unit. (2) Action Unit to emotion mapping - we developed a highly accurate expert network that sees through fake emotions and detects even the slightest micro-expressions2. This results in a complete and accurate facial emotion recognition system.

1An Action Unit [AU] is a facial muscle contraction.
2Micro-expressions are discussed in Section 2.1

52 pages

Thesis Committee:
Khaled Harras (Advisor, CMUQatar)
Bhiksha Raj (Co-Advisor)
Rita Singh (Co-Advisor)
Jeffery Cohon (University of Pittsburgh)

Srinivasan Seshan, Head, Computer Science Department
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science

This page maintained by