Real-Time Micro-Expression Recognition Using YOLOv8 and FER2013 Dataset

Abstract

Facial micro-expression recognition plays a crucial role in affective computing, human-computer interaction, and psychological analysis. This study implements a real-time face emotion detection system using the YOLOv8 (You Only Look Once version 8) model. The proposed approach leverages YOLO’s real-time processing capability and deep learning-based feature extraction to detect subtle facial muscle movements with high precision. The methodology involves pre-processing the FER2013 (Facial Expression Recognition 2013) dataset, which consists of grayscale images categorized into seven emotional classes.  The system can classify facial emotions into seven categories: Angry, Disgust, Fear, Happy, Sad, Surprise, and Neutral. Data augmentation techniques are employed to enhance model generalization, followed by training a custom YOLO model adapted for micro-expression recognition. Experimental results indicate that YOLO when fine-tuned with appropriate anchor boxes and a specialized loss function, can effectively classify facial micro-expressions. The study also discusses the impact of various hyperparameters and transfer learning techniques on performance. The findings demonstrate the potential of YOLO in real-time emotion recognition applications, paving the way for advancements in automated emotion analysis, security, and human behavior research.