A Comprehensive Review of Microexpression Recognition, Classification, and Datasets

Abstract

Facial microexpressions (MEs) are brief, involuntary facial movements that reveal genuine emotions a person attempts to suppress or conceal. Their short duration and low intensity pose considerable challenges for human observers and automated recognition systems, yet MEs hold practical value in psychotherapy, deception detection, marketing, and public safety. Several surveys have reviewed ME recognition, but most are primarily descriptive and lack systematic frameworks for literature selection and analysis. They also tend to overlook how individual, expression-related, and contextual factors shape recognition outcomes. This study addresses these limitations through a preferred reporting items for systematic reviews and meta-analyses (PRISMA)-based systematic review of ME recognition research published between 2015 and 2025, covering preprocessing methods, feature representation, and dataset characteristics. The review critically compares existing approaches, weighing their strengths, limitations, and suitability for real-world deployment. It examines persistent challenges, including dataset imbalance, limited cross-dataset generalization, and underexplored temporal dynamics, alongside emerging directions in action unit-based, multimodal, and deep learning methods. Specific recommendations for future work aimed at closing the gap between laboratory findings and practical application are also provided. By grounding the analysis in transparent selection criteria and structured synthesis, this work offers a reproducible foundation for advancing ME recognition research.