Facial Emotion Recognition: Leveraging Transfer Learning for Enhanced Decoding

Abstract

The recognition of facial emotions, also known as facial emotion recognition (FER), continues to be a crucial component of human-computer interaction and artificial intelligence applications. Various approaches are being developed to improve the accuracy of FER. This research article provides a thorough examination of Facial Expression Recognition (FER), with a specific emphasis on doing a comparison study between Convolutional Neural Networks (CNN) and Transfer Learning models. The primary objective is to get the highest possible accuracy in the categorization of emotions. This paper thoroughly investigates the complex terrain of facial expression recognition (FER), while recognising the difficulties presented by diverse lighting conditions, face emotions, and subtleties within the dataset. This study explores the capabilities of Convolutional Neural Networks (CNN), a well- established deep learning architecture, and Transfer Learning, a technique that utilises pre-trained models, in effectively capturing the nuanced aspects of facial expressions. The experimentation include rigorous training and testing on a wide range of datasets, assessing the accuracy, robustness, and generalizability of the models across many situations. Both CNN and Transfer Learning offer impressive accuracy in the field of Facial Expression Recognition (FER), with each approach showcasing distinct capabilities in addressing certain issues. Furthermore, this study examines the interpretability of judgements made by both models, providing insights into the facial areas that have a substantial impact on the results of emotion identification. The present research offers significant insights into the underlying mechanisms of these models, hence enhancing our comprehension of their effectiveness in practical contexts.