Abstract

In the domains of computer vision and artificial intelligence, the facial expression based emotion recognition is increasingly recognizes as a key research area and industrial challenge. Its increasing significance is reflected in diverse applications such as interactive gaming, intelligent home systems, expression and gesture analysis, surveillance monitoring, mental health therapy, patient care, and stress or anxiety assessment. This study examines specifically on studies that utilize facial images for Facial Expression Recognition (FER), since expressions represent a primary channel of non-verbal communication among humans. The quick evolution of deep learning has led to widespread adoption of its architectures, greatly improving recognition accuracy and efficiency. We examine machine learning, deep learning, and hybrid techniques in FER, including their use of preprocessing, data augmentation, and feature extraction techniques for analyzing temporal patterns across sequential frames. Additionally, the study provide an overview of widely used evaluation protocols and compare benchmark results, which serve as reliable measures for assessing FER performance. This survey is intended to serve both newcomers seeking foundational knowledge in FER and experienced researchers exploring new directions, offering a broad understanding of the cutting edge developments in the domain.

Key words: Facial Expression Recognition (FER); Deep Learning; Hybrid Models; Temporal Feature Extraction; Emotion Classification.