Deep Learning Based Approaches for Facial Emotion Recognition Using CNN Architectures: A Review

With the rise in Digital Communication, the ability to accurately recognize facial emotions must be improved to enhance virtual interpersonal interaction and many applications like affective computing, social robotics, mental health, and human–computer interaction. Convolutional neural networks (CNNs) are a powerful technique for Facial Emotion Recognition (FER) that can learn features from raw images and achieve high accuracy. However, many factors affect the performance of CNNs for FER, such as the choice of network architecture, the input representation, and the image pre-processing techniques. This paper reviews the recent advances in FER using CNNs and compares different approaches for each factor by focusing on several aspects: like datasets associated with FER, and the use of different image pre-processing techniques, use of models whether it is transfer learning or custom-based CNN models. This study essentially provides a thorough overview of FER using CNN, providing insights into its evolution, current landscape, challenges, and its potential to revolutionize human–computer interaction across various domains.