Optimized Facial Emotion Recognition Technique for Assessing User Experience Conference Paper

abstract

• This paper presents a novel optimization technique in image processing for emotion recognition based on facial expression. The method combines two pre-processing filters (pre-filters), i.e., brightness and contract filter and edge extraction filter, with Convolutional Neural Network (CNN) based learning and Support Vector Machine (SVM) for emotion classification. Instead of using an arbitrarily selected set of parameters used for the pre-filters, the presented algorithm automatically tunes the parameters by analyzing learning outcomes from CNN and selecting the parameter set which produces the best result. This method was evaluated for accuracy and efficiency. The result showed 98.19% emotion recognition accuracy using CNN with 3,500 epochs for given 3,589 face images. With demonstrated efficiency and accuracy, this method shows great potential for embedded human computer applications, in particular for assessing user experience and preference in games and media.

authors

• Lee, Kiju

published proceedings

• 2018 IEEE Games, Entertainment, Media Conference (GEM)

author list (cited authors)

• Liu, X., & Lee, K.

citation count

• 5

complete list of authors

• Liu, Xiao||Lee, Kiju

publication date

• August 2018

keywords

• Convolutional Neural Network
• Emotion Recognition
• Image Processing

Digital Object Identifier (DOI)

• 10.1109/GEM.2018.8516518

start page

• 407

end page

• 413

URL

• https://ieeexplore.ieee.org/document/8516518