Classification of the emotional stress and physical stress using signal magnification and canonical correlation analysis

• Published in: Pattern recognition Vol. 77; pp. 140 - 149
• Main Authors: Hong, Kan, Liu, Guodong, Chen, Wentao, Hong, Sheng
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.05.2018
• Subjects: Emotion classification; Emotional stress; Physical stress; Physical stress; Emotion classification; Emotional stress
• ISSN: 0031-3203; 1873-5142
• DOI: 10.1016/j.patcog.2017.12.013

•The emotional state of a person can be obtained by extracting individual stress index, which provides useful information to health field.•This paper directs the frontier of industrializing stress recognition by focusing on establishing a set of non-contact imaging based classifications for emotional stress (ES) and Physical stress (PS).•The proposed algorithm successfully extracts weak signal from thermal imaging for stress classification.•The classification algorithm is effective with accuracy as high as 90%. This study can lead to a practical system for the noninvasive assessment of stress states for practical applications. In affective computing, stress recognition mainly focuses on the relation of stress and photoelectric information. Researchers have used artificial intelligence to determine stress and computer identification channels. However, in applications such as health and security, Emotional stress (ES) information is usually to be alongside physical stress (PS) information, making it urgent to classify ES and PS. The thermal signals of ES and PS have yet to be classified, for which, signal amplification is offered. In this study, we propose a classification algorithm based on signal amplification and correlation analysis called Eulerian magnification-canonical correlation analysis. This signal amplification algorithm expands the signals of ES and PS in different frequency domains. Sparse coding and canonical correlation analysis then fuse the original signal and its amplified features. The extracted entropy features are used to train the correlation weight between ES and PS, which formulates stress classifications. The new classification method achieves an accuracy rate of 90%. This study can lead to a practical system for the noninvasive assessment of stress states for health or security applications.