Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Xiao_submission_ohiolink.pdf (10.39 MB)

ETD Abstract Container

Abstract Header

AUTOMATED FACIAL EMOTION RECOGNITION: DEVELOPMENT AND APPLICATION TO HUMAN-ROBOT INTERACTION

Liu, Xiao
http://rave.ohiolink.edu/etdc/view?acc_num=case1559936629839372

Abstract Details

, Master of Sciences, Case Western Reserve University, EMC - Mechanical Engineering.
This thesis presents two image processing algorithms for facial emotion recognition (FER). The first method uses two pre-processing filters (pre-filters), i.e., brightness and contrast filter and edge extraction filter, combined with Convolutional Neural Network (CNN) and Support Vector Machine (SVM). By using optimal pre-filter parameters in the pre-processing of the training images, the classification of FER could reach 98.19\% accuracy using CNN with 3,500 epochs for 3,589 face images from the FER2013 datasets. The second approach introduces two geometrical facial features based on action units -- landmark curvatures and vectorized landmarks. This method first detects facial landmarks and extracts action unit (AU) features. The extracted facial segments based on the action units are classified into five groups and input to a SVM. The presented method show how individual parameters, including detected landmarks, AU group selection, and parameters used in the SVM, can be examined and systematically selected for the optimal performance in FER. The results after parameter optimization showed 98.38\% test accuracy with training using 1,479 labeled frames of Cohn-Kanade (CK+) database, and 98.11\% test accuracy with training using 1,710 labeled frames of Multimedia Understanding Group (MUG) database for 6-emotion classification. This technique also shows the real-time processing speed of 6.67 frames per second (fps) for images with a 640x480 resolution. The novelty of the first approach is combining image processing filters with CNN to enhance CNN performance. As for the second approach, it systematically analyzed the effectiveness of proposed geometric features and implemented FER in real-time. The demonstrated algorithms have been applied on human-robot interaction (HRI) application platform - social robot ``Woody" for testing. The presented algorithms have been made publicly available.
Kiju Lee (Committee Chair)
Kathryn Daltorio (Committee Member)
Frank Merat (Committee Member)
Computer Science; Mechanical Engineering; Robotics
Facial Emotion Recognition, Deep Learning, Human-robot Interaction

Recommended Citations

Citations

  • Liu, X. (n.d.). AUTOMATED FACIAL EMOTION RECOGNITION: DEVELOPMENT AND APPLICATION TO HUMAN-ROBOT INTERACTION [Master's thesis, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1559936629839372

    APA Style (7th edition)

  • Liu, Xiao. AUTOMATED FACIAL EMOTION RECOGNITION: DEVELOPMENT AND APPLICATION TO HUMAN-ROBOT INTERACTION . Case Western Reserve University, Master's thesis. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=case1559936629839372.

    MLA Style (8th edition)

  • Liu, Xiao. "AUTOMATED FACIAL EMOTION RECOGNITION: DEVELOPMENT AND APPLICATION TO HUMAN-ROBOT INTERACTION ." Master's thesis, Case Western Reserve University. Accessed APRIL 25, 2024. http://rave.ohiolink.edu/etdc/view?acc_num=case1559936629839372

    Chicago Manual of Style (17th edition)

case1559936629839372
266
© , all rights reserved.
This open access ETD is published by Case Western Reserve University School of Graduate Studies and OhioLINK.