Identifying emotions from facial expressions using a deep convolutional neural network-based approach

• Published in: Multimedia tools and applications Vol. 83; no. 6; pp. 15711 - 15732
• Main Authors: Meena, Gaurav, Mohbey, Krishna Kumar, Indian, Ajay, Khan, Mohammad Zubair, Kumar, Sunil
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.02.2024; Springer Nature B.V
• Subjects: Artificial neural networks; Computer Communication Networks; Computer Science; Data mining; Data Structures and Information Theory; Datasets; Face recognition; Machine learning; Man-machine interfaces; Multimedia Information Systems; Neural networks; Sentiment analysis; Special Purpose and Application-Based Systems; Deep learning; Convolutional neural networks; Facial sentiment analysis
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-16174-3

Sentiment identification on facial expression is an interesting study domain with applications in various disciplines, including security, health, and human-machine interfaces. The main goal of sentiment analysis is to decide an individual’s perspective on a topic or the document’s overall contextual polarity. In nonverbal communication, sentiment analysis plays a vital role in an individual’s feelings, reflecting on the faces. Researchers in this area are interested in improving models and methods and extracting various characteristics to provide a better computer prediction of sentiments. Sentiment polarities are mainly classified as positive, negative, and neutral. Many sentiment analysis approaches exist, but deep learning architectures can handle extensive data and provide better performances. We presented a solution based on the CNN (Convolutional Neural Network) model for handling this problem. This work uses the extended Cohn Kanade (CK+) and FER-2013 datasets for facial expression recognition study. Several existing architectures are used to evaluate the efficiency of the proposed model. Extensive experiments are carried out on both CK+ and FER-2013 data sets, and our framework outperforms state-of-the-art techniques. According to obtained results, the CNN3 model gives 79% and 95% accuracy for FER-2013 and CK+ datasets, respectively.