Adaptive 3D facial action intensity estimation and emotion recognition

• Published in: Expert systems with applications Vol. 42; no. 3; pp. 1446 - 1464
• Main Authors: Zhang, Yang, Zhang, Li, Hossain, M.A.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier Ltd 15.02.2015; Elsevier
• Subjects: Action unit intensity estimation; Adaptive ensemble classifiers; Anatomical correlates of behavior; Applied sciences; Artificial intelligence; Behavioral psychophysiology; Biological and medical sciences; Complementary Neural Networks; Computer science; control theory; systems; Connectionism. Neural networks; Dynamical systems; Dynamics; Emotions; Exact sciences and technology; Face recognition; Facial emotion recognition; Feature recognition; Fundamental and applied biological sciences. Psychology; Ground, air and sea transportation, marine construction; Pattern recognition. Digital image processing. Computational geometry; Psychology. Psychoanalysis. Psychiatry; Psychology. Psychophysiology; Real time; Road transportation and traffic; Support vector classification; Support Vector Regression; Tasks; Three dimensional; Action unit intensity estimation; Support vector classification; Complementary Neural Networks; Adaptive ensemble classifiers; Facial emotion recognition; Support Vector Regression; Action; Customization; Adaptive estimation; Flexibility; Modeling; Adaptive method; Optimization; Aggregate model; Vehicle driver; Uncertain system; Feedforward neural nets; Behavioral analysis; Database; Vector support machine; Illumination; Feedforward; Monitoring; Psychophysiology; Computer vision; Head; Motion estimation; Face recognition; Redundancy; Cluster; Emotion emotionality; Regression analysis; Neural network; Real time; Luminance; Relevance criterion; Surveillance; Scene analysis; Body movement; Automatic recognition; Facial expression

•We estimate AU intensity based on mRMR-based feature selection.•Adaptive ensemble classifiers are proposed for robust expression classification.•The ensemble models show superior ability for novel emotion class detection.•Both database images and real human subjects are used to evaluate the system....