Roy Arnab Kumar ; Kathania Hemant Kumar ; Sharma Adhitiya ; Dey Abhishek ; Ansari Md. Sarfaraj Alam

Abstract

The human face is a silent communicator, expressing emotions and thoughtsthrough its facial expressions. With the advancements in computer vision inrecent years, facial emotion recognition technology has made significantstrides, enabling machines to decode the intricacies of facial cues. In thiswork, we propose ResEmoteNet, a novel deep learning architecture for facialemotion recognition designed with the combination of Convolutional,Squeeze-Excitation (SE) and Residual Networks. The inclusion of SE blockselectively focuses on the important features of the human face, enhances thefeature representation and suppresses the less relevant ones. This helps inreducing the loss and enhancing the overall model performance. We alsointegrate the SE block with three residual blocks that help in learning morecomplex representation of the data through deeper layers. We evaluatedResEmoteNet on four open-source databases: FER2013, RAF-DB, AffectNet-7 andExpW, achieving accuracies of 79.79%, 94.76%, 72.39% and 75.67% respectively.The proposed network outperforms state-of-the-art models across all fourdatabases. The source code for ResEmoteNet is available athttps://github.com/ArnabKumarRoy02/ResEmoteNet.