Facial Expression Recognition in the Wild: A Comprehensive Study on Datasets, Methods, and Challenges

Considerable progress has been made in controlled laboratory environments regarding Automatic Facial Expression Recognition (AFER); however, the implementation of AFER in real-world, unconstrained settings poses an array of challenges. Recent studies on AFER systems have concentrated on prominent issues, including overfitting due to a lack of training data, generalizability, resilience to identity bias, lighting, head position variation, partial occlusion, and crossdataset experiments. This paper explores the field of facial expression recognition in the wild (FERW) and provides an indepth analysis of the issues, solutions, and difficulties that arise in real-world situations. The benchmark datasets for face expression identification in the wild settings and their characteristics are summarized. The state-of-the-art methods to derive the highly discriminative and abstract feature distributions that offer robustness to unconstrained environments are discussed. The methodologies used to evaluate FERW systems' performance are outlined, with a comparative analysis of the approaches applied to benchmark datasets. In addition, challenging unresolved issues in the field of FERW are described in depth, thereby establishing an open-ended research domain pertaining to the problem.