Static Facial Expression Recognition in the Wild: Taxonomy, Trends and Challenges

Abstract

In recent years, Facial Expression Recognition (FER) has gained significant attention due to its wide application and potential in various domains. FER is the research field that focuses on recognizing and classifying human emotions expressed by humans into emotion categories using computer vision. Different machine learning techniques have been applied to this research field with promising outcomes through the application of increasingly more powerful machine learning algorithms. This systematic literature review is conducted to investigate static FER on unconstrained datasets. A total of 32 studies were retrieved from four major academic repositories. The aim of this study is to provide a comprehensive review of static FER research on unconstrained facial expression image datasets including the overview of key concepts, the approaches applied, the datasets used, the current state-of-the-art as well as the future directions of research in this fast-developing research field. Deep learning methods emerged as the most promising approach for static FER while second-order pooling in CNNs allowed for improved representation of regional features and facial landmark distortion.