JP Journal of Heat and Mass Transfer
Special Issue, Advances in ICT-Convergence, Pages 71 - 81
(September 2020) http://dx.doi.org/10.17654/HMSI20071 |
|
LIGHTWEIGHT DEEP FACIAL EXPRESSION RECOGNITION USING GEOMETRIC FEATURES
Jinho Han
|
Abstract: Deep facial expression recognition (FER) involves determining facial expressions using deep learning techniques, including deep neural networks. In 2012, AlexNet used deep learning with a technique called convolutional neural networks for facial recognition (FR). Since then, more than 99% accuracy has been achieved in FR using deep learning-based approaches. Deep neural networks require millions of parameters; that is, millions of weights and biases that require very high levels of computational performance to achieve high accuracy in FR. These kinds of techniques cannot be applied to mobile devices, which require lightweight applications. In this paper, we propose a lightweight, deep learning-based FER method for mobile devices. In an experiment using the Yale Face Database, our proposed method using a deep FER system achieved 86.7% accuracy and our model size was only 80KB, which equates to 0.0003% of AlexNet. |
Keywords and phrases: facial expression recognition, mobile devices, lightweight deep learning technique.
|
|
Number of Downloads: 267 | Number of Views: 1589 |
|