Facial Expression Recognition Using Convolutional Neural Networks and Stochastic Gradient Descent Optimization Algorithms
Keywords:
Facial Expression Recognition (FER), Stochastic Gradient Descent (SGD), Deep Convolutional Neural Networks (DCNNs), Root Mean Squared Propagation (RMSprop)Abstract
This paper presents the design of a Facial Expression Recognition (FER) system using Deep Convolutional Neural Networks (DCNNs) to accurately identify seven key human facial expressions. The DCNN module and FER system were trained and tested on various facial datasets, including JAFFE, KDEF, MUG, WSEFEP, ADFES, and TFEID. The experiments involved testing different models and architectures with varying numbers of convolutional layers, filter sizes, and epochs. Results from these experiments are based on 2982 images of faces on the seven-basic expression. The study also evaluated the performance of Stochastic Gradient Descent (SGD), Root Mean Squared Propagation (RMSprop), and Adaptive Moment Estimation (Adam), optimization algorithms on the DCNN architecture. Results showed that SGDM with an adaptive learning-rate achieved the highest validation accuracy of 98.35%, outperforming other algorithms. Additionally, the study found that RMSprop led to unstable training and lower accuracy, while Adam did not significantly improve accuracy with adaptive learning rate.
The research demonstrated that selecting the right combination of model elements led to improved accuracy and convergence time. The system achieved a recognition rate of 98.35% for the tested dataset using the DCNN algorithm, highlighting its effectiveness in facial expression recognition.
Published
How to Cite
Issue
Section
![Creative Commons License](http://i.creativecommons.org/l/by/4.0/88x31.png)
This work is licensed under a Creative Commons Attribution 4.0 International License.