Micro-expression spotting based on supervised learning

Micro-expressions are fast and involuntary facial expressions, which indicate concealed or suppressed emotions. Recently, micro-expression analysis has become an attractive topic. There have been many studies involved in this problem due to potential applications for assisting human-centered interactions and communication. However, there are still limitations, especially in micro-expression spotting. This is one task of micro-expression analysis for locating the temporal position of micro-expressions in video sequence.

This thesis is a thorough summary of the main topic which deals with exploring supervised learning for micro-expression spotting consisting of five publications. Firstly, the data preparation and benchmarking for micro-expressions spotting are presented, including video frame selection, face image alignment, and new benchmarks for micro-expressions spotting to standardize the evaluation of micro-expressions spotting. Secondly, machine learning models for facial expression analysis and micro-expressions spotting are introduced, from handcrafted features to deep learning techniques.

The contributions of this research are as follows. Firstly, I extend a spontaneous micro-expressions database to create a larger version to achieve better micro-expression spotting tasks. Additionally, I introduce new benchmarks for micro-expressions spotting to obtain a fair comparison for micro-expressions spotting techniques. Moreover, I introduce our findings to a new method based on the multi-modal approach to spot micro-expression in long videos. Finally, I summarize the contributions of the work and discuss future research.