Subject-independent emotion recognition based on physiological signals: a three-stage decision method

Area of Science:

• Computer Science
• Affective Computing
• Biomedical Engineering

Background:

• Human-computer interaction systems lack emotional understanding.
• Physiological signals correlate with emotional states, offering a channel for emotion detection.
• Accurate emotion recognition is key for responsive human-computer systems.

Purpose of the Study:

• To propose a novel three-stage decision method for recognizing four emotions from physiological signals.
• To address the challenge of individual differences in multi-subject emotion recognition.
• To improve the accuracy and robustness of emotion detection in human-computer interaction.

Main Methods:

• A three-stage decision process involving subject grouping, emotion pool categorization, and classifier training.
• Eliminating individual differences by transforming mixed training subjects into separate groups in the initial stage.
• Reducing recognition complexity by categorizing four emotions into two emotion pools in the second stage.

Main Results:

• Achieved an average recognition accuracy of 77.57% for four emotions, with a peak accuracy of 86.67% for positive and excited emotions.
• Demonstrated significant improvement compared to other methods in multi-subject emotion recognition.
• Identified variations in classifier effectiveness across different emotion allocations.

Conclusions:

• The proposed three-stage method effectively addresses 'individual differences' in multi-subject emotion recognition.
• The methodology overcomes suboptimal performance associated with direct multi-emotion classification.
• This approach represents a promising methodology for recognizing multiple emotions in diverse, multi-subject scenarios.