An Efficient Deep Learning based Real Time Facial Expression Recognition System

Abstract

Human-computer interaction, emotion analysis, and more depend on computer vision's face expression recognition. Real-time facial expression identification enhances decision-making in human-computer interaction, healthcare, marketing, and other fields. This study introduces a fast and accurate real-time facial expression identification system utilizing Convolutional Neural Network (CNN) and Mobile Network Version 2 (MobileNetV2). The proposed methodology involves the design and implementation of a CNN architecture tailored for facial expression recognition. We strategically arrange layers, fine-tune parameters, and leverage transfer learning techniques, particularly focusing on the optimization of model depth and complexity. The model is trained and evaluated using benchmark datasets to ensure robust performance across various facial expressions. Data-driven infrastructure helps the system manage real-world changes. Our approach identifies facial emotions in real time from live footage. This paper explores facial expression recognition employing a simplified CNN architecture, contrasting with more complex pre-trained networks like MobileNetV2. Despite its simplicity, the proposed CNN yields result close to the performance of MobileNetV2. The study emphasizes the viability of less intricate CNNs for facial expression recognition, offering a balance between model simplicity and competitive accuracy. High accuracy and computational efficiency make the system suitable for real-time applications and resource-constrained systems The experimental tests extracted that a very acceptable accuracy achieved when using CNN model with the AffectNet dataset with a ratio of (97.2%).