Innovative Facial Detection and Emotion Recognition System Utilizing Correlation Attention Module in Deep Convolutional Neural Networks

Abstract

Facial detection and emotion recognition are pivotal in human-computer interaction, facing numerous challenges such as varying illuminations and occlusions. Despite efforts by multiple researchers, labeling diverse emotions remains a challenge, necessitating improvements in existing models. This study aims to introduce a deep learning-based facial emotion detection model capable of accurately discerning emotions from images. Leveraging a deep Convolutional Neural Network (CNN) classifier, facial expressions are identified with precision, thanks to effective feature learning from images. Additionally, a correlation attention module is devised to enhance the classifier's efficacy by establishing relationships between features extracted by Residual Network 101 (ResNet 101) and VGG 16. Evaluated on CK+48 and Japanese Female Facial Expression (JAFFE) datasets, the face detection model achieves efficiencies of 94.46%, 95.00%, and 92.19%, while the emotion detection model scores 98.73%, 98.10%, and 99.55%, and 98.73%, 98.10%, and 98.55% respectively, in terms of accuracy, sensitivity, and specificity.