Machine learning for perceiving facial micro-expression

Thesis event information

Date and time of the thesis defence

Place of the thesis defence

TA 105, Linnanmaa

Topic of the dissertation

Machine learning for perceiving facial micro-expression

Doctoral candidate

Master of Science Yante Li

Faculty and unit

University of Oulu Graduate School, Faculty of Information Technology and Electrical Engineering, Center for Machine Vision and Signal Analysis (CMVS)

Subject of study

Computer Science and Engineering

Opponent

Professor Robert Jenssen, UiT The Arctic University of Norway

Custos

Academy Professor Guoying Zhao, University of Oulu

Visit thesis event

Add event to calendar

Machine learning for perceiving facial micro-expression

Emotion analysis plays an important role in humans' daily lives. Facial expression is one of the major ways to express emotions. Besides the common facial expressions we see every day, emotion can also be expressed in a special format, micro-expression. Micro-expressions (MEs) are involuntary facial movements that come about in reaction to emotional stimulus, which reveal people’s hidden feelings in high-stakes situations and have many potential applications, such as clinical diagnosis, ensuring national security, and conducting interrogations. However, ME recognition becomes challenging due to the low intensity, short duration and small-scale datasets.

This thesis is a through summary of the important subjects for ME recognition, consisting of five papers corresponding to the progress of my research. Firstly, the automatic ME recognition system based on deep learning is introduced. Secondly, the Micro-expression Action Unit (ME-AU) detection is described, which plays an important role in facial behavior analysis. Thirdly, the robust ME recognition with AU detection is illustrated that verifies the contribution of AU detection to ME recognition.

The contributions of this study can be classified into three categories: (1) A deep ME recognition approach with the apex frame is proposed, which would be capable of demonstrating that deep learning can achieve impressive performance of ME recognition with the apex frame; (2) We break the ground of the ME-AU study and provide the baselines and novel transfer learning methods for the future study of ME-AU detection; (3) A unified framework for ME recognition with AU detection based on contrastive learning is proposed for verifying the AU contribution to robust ME recognition.

Lastly, we summarize the contributions of the work, and propose future plans about ME studies based on the limitations of the current work.
Last updated: 23.1.2024