A CNN Model for Facial Emotion Recognition

Abstract

Aims: To develop and evaluate a computational technique for identification and categorization of human emotion based on facial expression, automatically. Study Design: This study highlights deep learning's role in enhancing facial expression recognition and suggests future advancements for real-time applications. A deep neural network (DNN) for facial emotion recognition (FER) using a combination of convolution neural networks (CNN), squeeze-and-excitation networks, and residual neural networks were used to identify critical facial features for FER, focusing on areas around the nose and mouth. The study utilized AfectNet and the Real-World Affective Faces Database (RAF-DB) for training. Place and Duration of Study: Department of Computer Science and Engineering (CSE), Technocrats Institute of Technology (TIT) College, Bhopal (MP), between May 2024 and February 2025. Methodology: A deep learning approach using the VGG-16 model was employed and started with dataset loading and pre-processing through an image Data store, and ensuring class balance by splitting the dataset evenly before dividing it into 70% training and 30% testing sets. Images are resized for VGG-16 input, and grayscale images are converted to RGB. Performance evaluation utilizes a confusion matrix to measure accuracy, sensitivity, specificity, precision, recall, Jaccard coefficient, and Dice coefficient. Results: The model achieves a high Sensitivity (95.45%) and Specificity (94.69%), indicating its ability to correctly classify positive and negative instances. The Precision and Recall values are both 94.69%, reflecting the model’s balance in identifying relevant instances. The existing SVM-based system achieves an accuracy of 83.01%, whereas the proposed VGG16 model significantly improves accuracy to 95.45%. Conclusion:The study showcases the VGG16 model's effectiveness for facial expression recognition and strong metrics in terms of sensitivity, specificity, precision, and recall.