Bodapati, Jyostna Devi
FERNet: a deep CNN Architecture for facial expression recognition in the wild - Vol.103(2), Apr - New York Springer 2022 - 439-448p.
Facial expression recognition is an intriguing and demanding subject in the realm of computer vision. In this paper, we propose a novel deep learning-based strategy to address the challenges of facial expression recognition from images. Our model is developed in such a manner that it learns hidden nonlinearity from the input facial images, which is critical for discriminating the type of emotion a person is expressing. We developed a deep convolutional neural network model composed of a sequence of blocks, each consists of multiple convolutional layers and sub-sampling layers. Investigations on the benchmark FER2013 dataset indicate that the proposed facial expression recognition network (FERNet) surpasses existing approaches in terms of performance and model complexity. We trained our model on the FER2013 dataset, which is the most challenging of all the available datasets for this task, and achieve an accuracy of around 69.57%. Furthermore, we investigate the effects of dropout, batch normalization, and augmentation, as well as how they aid in the reduction of over-fitting and improved performance.
Humanities and Applied Sciences
FERNet: a deep CNN Architecture for facial expression recognition in the wild - Vol.103(2), Apr - New York Springer 2022 - 439-448p.
Facial expression recognition is an intriguing and demanding subject in the realm of computer vision. In this paper, we propose a novel deep learning-based strategy to address the challenges of facial expression recognition from images. Our model is developed in such a manner that it learns hidden nonlinearity from the input facial images, which is critical for discriminating the type of emotion a person is expressing. We developed a deep convolutional neural network model composed of a sequence of blocks, each consists of multiple convolutional layers and sub-sampling layers. Investigations on the benchmark FER2013 dataset indicate that the proposed facial expression recognition network (FERNet) surpasses existing approaches in terms of performance and model complexity. We trained our model on the FER2013 dataset, which is the most challenging of all the available datasets for this task, and achieve an accuracy of around 69.57%. Furthermore, we investigate the effects of dropout, batch normalization, and augmentation, as well as how they aid in the reduction of over-fitting and improved performance.
Humanities and Applied Sciences