Comparative Assessment of Facial Expression Recognition Models for Unraveling Emotional Signals with Convolutional Neural Networks

Facial expressions are vital in conveying human emotions, forming a foundation of non-verbal communication. Recognizing these expressions computationally known as Facial Expression Recognition (FER) holds great significance in Artificial Intelligence (AI) research. This study focuses on employing Convolutional Neural Networks (CNNs) for FER classification using static images, omitting pre-processing or feature extraction. Our approach integrates pre-processing steps like face detection and illumination correction to bolster future accuracy. Through feature extraction, we pinpoint critical facial areas jaw, mouth, eyes, nose, and eyebrows enhancing computational modelling. Additionally, we survey existing literature to inform our CNN architecture, addressing challenges posed by components like max-pooling and dropout layers, culminating in improved performance. Our experiments attained a 78.9%test accuracy in the seven-class FER2013 dataset classification and 82.03% in FER plus dataset. This work not only advances technology but also lays the groundwork for further decoding the nuanced language of facial expressions.