Facial Micro-Expression Recognition (FMER) using Model Compression

Abstract

Emotion recognition that utilizes facial expressions is critical for AI systems like social robots to interact effectively with humans. In the actual world, however, it is considerably more difficult to distinguish face micro-expressions (FMEs) than it is to recognize facial general-expressions with complex emotions. It’s a form of facial expression that lasts only a few seconds, has a modest magnitude, and usually local movement. However, because of these properties of MEs (micro-expressions), obtaining ME data is challenging, which is a constraint when using deep learning algorithms to recognize FMEs. The facial action coding system (FACS) can likewise encode these FMEs which represents that there is some relation between Action Unit (AU) and FMEs. To demonstrate this, a knowledge transfer scheme for ME recognition that compress and transfers information from an AU is given in this manuscript. Experiments are carried out on four publicly available ME database. Our model outperforms state-of-the-art systems, according on the findings of the experiments.