TY - GEN
T1 - A comparison of facial features and fusion methods for emotion recognition
AU - Smirnov, Demiyan V.
AU - Muraleedharan, Rajani
AU - Ramachandran, Ravi P.
N1 - Funding Information:
This work was supported by the National Science Foundation through Grant DUE-1122296.
Publisher Copyright:
© Springer International Publishing Switzerland 2015.
PY - 2015
Y1 - 2015
N2 - Emotion recognition is an important part of human behavior analysis. It finds many applications including human-computer interaction, driver safety, health care, stress detection, psychological analysis, forensics, law enforcement and customer care. The focus of this paper is to use a pattern recognition framework based on facial expression features and two classifiers (linear discriminant analysis and k-nearest neighbor) for emotion recognition. The extended Cohn-Kanade data- base is used to classify 5 emotions, namely, ‘neutral, angry, disgust, happy, and surprise’. The Discrete Cosine Transform (DCT), Discrete Sine Transform (DST), the Walsh-Hadamard Transform (FWHT) and a new 7-dimensional feature based on condensing the Facial Action Coding System (FACS) are compared. Ensemble systems using decision level, score fusion and Borda count are also studied. Fusion of the four features leads to slightly more than a 90% accuracy.
AB - Emotion recognition is an important part of human behavior analysis. It finds many applications including human-computer interaction, driver safety, health care, stress detection, psychological analysis, forensics, law enforcement and customer care. The focus of this paper is to use a pattern recognition framework based on facial expression features and two classifiers (linear discriminant analysis and k-nearest neighbor) for emotion recognition. The extended Cohn-Kanade data- base is used to classify 5 emotions, namely, ‘neutral, angry, disgust, happy, and surprise’. The Discrete Cosine Transform (DCT), Discrete Sine Transform (DST), the Walsh-Hadamard Transform (FWHT) and a new 7-dimensional feature based on condensing the Facial Action Coding System (FACS) are compared. Ensemble systems using decision level, score fusion and Borda count are also studied. Fusion of the four features leads to slightly more than a 90% accuracy.
UR - http://www.scopus.com/inward/record.url?scp=84951876645&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84951876645&partnerID=8YFLogxK
U2 - 10.1007/978-3-319-26561-2_68
DO - 10.1007/978-3-319-26561-2_68
M3 - Conference contribution
AN - SCOPUS:84951876645
SN - 9783319265605
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 574
EP - 582
BT - Neural Information Processing - 22nd International Conference, ICONIP 2015, Proceedings
A2 - Arik, Sabri
A2 - Huang, Tingwen
A2 - Lai, Weng Kin
A2 - Liu, Qingshan
PB - Springer Verlag
T2 - 22nd International Conference on Neural Information Processing, ICONIP 2015
Y2 - 9 November 2015 through 12 November 2015
ER -