Recognizing facial emotion in real-time using MuWNet a novel deep learning network
Facial expression recognition (FER) is a branch of psychology that studies the classification of human emotions using facial expressions. Particularly, FER can be implemented in a vast array of applications, including education, online entertainment, and even essential fields involving human lives a...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Penerbit Universiti Kebangsaan Malaysia
2024
|
| Online Access: | http://journalarticle.ukm.my/23979/1/1%20-%2020.pdf http://journalarticle.ukm.my/23979/ https://www.ukm.my/apjitm |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Facial expression recognition (FER) is a branch of psychology that studies the classification of human emotions using facial expressions. Particularly, FER can be implemented in a vast array of applications, including education, online entertainment, and even essential fields involving human lives and behavior, such as medicine. There are seven universal facial expression categories: surprise, sadness, happiness, contempt, fear, anger, and neutrality. Recognizing all these facial expressions and predicting a person's present mood is a challenging problem for machines. Because of the nature of humans, this challenge presents itself to a computer in a more sophisticated manner. The main objective of this research was to construct a novel deep Convolutional Neural Network (CNN) for facial expression classification that can assist in extracting features from images to identify facial gestures and then apply it in real-time. Various neural network models and classification methods have been introduced in the past to reach cutting-edge accuracy in this industry. Separate studies have investigated the capabilities and effectiveness of CNN models in distinguishing human emotions on the FER2013 dataset. In this study, the proposed MuWNet model has been diversified with several types of layers, such as convolution layers, separable convolution layers, and residual blocks. In addition, applying hyperparameter tweaking to enhance progress. The results of two experiments that have been done on the MuWNet model indicate that the accuracy of the classification in the second experiment was 70.72%, with an increase of 0.14% over the first. Finally, these results appear to be competitive in the field of FER, and it can be stated that the proposed model contributed to the emergence of a classification system for facial expressions. |
|---|