TV-L1 Optical Flow based Hierarchical Transformer Network for Micro-Expression Recognition

Accurate interpretation of human emotions and micro-expressions is vital for advancing human-computer interaction, enhancing security systems, and supporting psychological research. Micro-expression analysis enables systems to detect subtle non-verbal cues, fostering empathetic and intuitive interactions in domains such as security, customer service, and mental health diagnostics. This study presents a novel approach to micro-expression recognition by integrating a Hierarchical Transformer Network (HTNet) with the TV-L1 optical flow algorithm. HTNet's hierarchical transformer architecture efficiently captures and processes temporal features, while the TV-L1 algorithm robustly detects subtle facial motion dynamics essential for micro-expression analysis. The proposed model is fine-tuned through hyperparameter optimization and evaluated using metrics like accuracy, Unweighted F1-score, Unweighted Average Recall, Matthews Correlation Coefficient, and Expected Calibration Error. Experiments conducted on the SAMM and CASME II datasets demonstrate the model's high precision, real-time capabilities, and adaptability to dynamic environments. This research bridges the gap between cutting-edge AI models and practical emotion recognition systems, offering valuable insights into human emotional subtleties. The findings highlight the transformative potential of AI in understanding and responding to human emotions, laying the foundation for future advancements in human-centric applications across diverse fields.