Exploring the Interplay between Facial Expression Recognition and Physical States

This paper suggests a new viewpoint in Facial Expression Recognition (FER), moving beyond conventional approaches focused on understanding human emotions to include also physical states expressions such as pain and effort. These expressions involve facial muscle activities that deviate from straightforward emotional expressions, often overlooked by existing datasets and classifiers that predominantly focus on emotional states. The study presented addresses inaccuracies in facial expression reporting when the input image corresponds to a physical state. By utilizing a pre-trained FER classifier on a specialized dataset, this research analyses the implications of lacking classifiers tailored for physical states. Critical issues in FER tasks are highlighted, revealing how datasets without physical states labels introduce bias and impact accuracy. We consider the UIBVFED Physical States dataset, a dataset featuring facial expressions of physical states, to be a significant contribution. This dataset addresses biased estimations in FER tasks and enhances the training of recognition systems, improving their suitability across diverse scenarios.