Facial Expression Recognition (FER) is the task of identifying human emotions from facial cues. Its origins lie in psychological research, particularly the work of Paul Ekman, who proposed universal basic emotions and developed the Facial Action Coding System (FACS). The task of FER became relevant in the context of Machine Learning (ML) too, with ML models evolving from shallow-learning algorithms based on feature-based methods (e.g., Local Binary Patterns with Support Vector Machines) to modern Deep Learning (DL) approaches. This thesis thus deals with FER, especially in the presence of artificial occlusions, meaning on images partially occluded in post-production. The main goal is comparing how different human individuals or different machines approach the task.