Title:	A REAL-TIME FACIAL EXPRESSION RECOGNITION FOR ONLINE EMBEDDED DEVICES REQUIRING USER’S EMOTIONAL INTERACTION
Author(s):	Mi Ra Jeong, Duyoung Heo, Jae-Yeal Nam and Byoung Chul Ko
ISBN:	978-989-8533-79-1
Editors:	Katherine Blashki and Yingcai Xiao
Year:	2018
Edition:	Single
Keywords:	Facial Expression Recognition, Emotion, FER, Low Specification System, Embedded System, Random Forest
Type:	Full Paper
First Page:	214
Last Page:	220
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	Currently, much of the automatic facial expression recognition (FER) research focuses on the use of deep-learning algorithms beyond classic pattern classifiers. However, deep-learning based FER approaches require more computing resources to analyze the large amounts of data and the massive number of parameters associated with this approach. Because of this demand on computing resources, deep-learning based FER approaches are not appropriate for real-time embedded applications with limited computing resources. To address this problem, we propose using hierarchical boosted random forest (BRF) classifier instead of deep network algorithms for various embedded devices. For FER, we first detect the face region and facial landmarks from input images and extract geometric feature descriptors considering the spatial position between landmarks. We then apply these feature vectors to proposed hierarchical BRF classifiers and classify facial expressions for the input image. To estimate the performance, we applied the proposed algorithm to the extended Cohn-Kanade (CK+) dataset. The accuracy was 91.6%. Through experiments, the proposed algorithm yielded similar performance compared to deep learning FER algorithms, with a significantly reduced processing cost. Our hierarchical BRF approach is more suitable for real-time embedded systems requiring user emotional interaction like virtual reality (VR) gaming.
	Go Back