Improved Micro-expression Recognition: An Apex Frame-Based Approach Feature Tracking and KLT

This research develops a real-time facial micro-expression recognition system, focusing on analyzing the onset and apex phases of micro-expression on the Spontaneous Activity and Micro-Movements (SAMM) dataset. Micro- expressions are very brief (0.04 - 0.2 seconds) facial muscle movements that often occur when a person is trying to hide emotions. The developed system aims to improve computation time efficiency and micro-expression recognition accuracy by optimizing feature extraction techniques and selecting more specific facial areas, including facial components such as eyebrows, eyes, and mouth. This research successfully improved the computation time efficiency by 51.96%, almost half the time required by the previous method. In addition, this study shows an increase in efficiency compared to previous studies, with an increase of 34.3% for SVM with Manual Sampling technique and 32.6% for MLP-Backpropagation. In the Random Sampling technique, SVM efficiency increased by 6.1%, but MLP-Backpropagation accuracy decreased by 4.8%. This method achieved 77.9% accuracy for MLP- Backpropagation, which is higher than the previous method. This research contributes to accelerating micro- expression recognition systems and improving accuracy, which opens opportunities for real-time emotion analysis applications such as lie detection or human behavior monitoring in a broader context.