Assessing Sequential Databases for Spontaneous and Posed Facial Expression Recognition

Abstract

Advancements in AI for recognizing facial expressions of emotion rely heavily on the quality of underlying data. We present a comparative analysis of sequential databases for spontaneous (real) and posed (fake) facial expressions, introducing a modular, metric-based framework for evaluating data quality. This framework allows for flexible selection and weighting of metrics, making it adaptable to a wide range of research needs. Applied to 13 databases, it identifies key characteristics of an ideal data set, particularly for AI systems that distinguish between spontaneous and posed facial expressions. Our findings offer practical solutions to optimize data quality, laying a foundation for ensuring high-quality data in future emotion recognition research.