A Proficient Approach Utilising High-Level Representations for Micro-Expression Spotting

Lok, Kevin (2025) A Proficient Approach Utilising High-Level Representations for Micro-Expression Spotting. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Lok Kevin_Full Text.pdf Restricted to Registered users only Download (2MB)

Abstract

Micro-expressions are brief, involuntary facial expressions that occur within a fraction of a second, often lasting only a few tenths of a second. These expressions can reveal a person’s genuine emotions when trying to suppress them. However, the micro-expression spotting has some challenges. The micro-expression’s short duration and subtle movements make them difficult to detect. Besides, distinguishing between images with and without micro-expressions is difficult due to their high visual similarity. Therefore, the enhanced version of the NxN regional Cubic-Local Binary Pattern is proposed as a solution to overcome these problems. This method involves a sum of squared differences to compute the feature difference. Also, it incorporates the mean absolute error and standard error to evaluate the model performance across the micro-expression datasets. The results demonstrate that the 2x2 configuration using an 8-bin histogram significantly improves accuracy in detecting micro-expressions, with a mean absolute error of 2.7961 and standard error of 0.2119, outperforming the original method. This approach enhances detection accuracy and offers greater reliability and consistency, contributing to a deeper understanding of human emotions.