Facial Emotion Recognition in Verbal Communication Based on Deep Learning

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Facial emotion recognition is complex due to variable human expressions.
• Deep learning models show promise but suffer performance degradation from suboptimal convolutional neural network (CNN) layer selection.
• Existing methods require improvement for robust emotion classification.

Purpose of the Study:

• To propose an efficient deep learning (DL) technique using an improved CNN architecture for facial emotion recognition.
• To address performance degradation issues in current DL models for emotion classification.
• To develop a reliable method for classifying emotion type and intensity from facial images.

Main Methods:

• Developed an improved CNN model architecture specifically designed for processing aggregated facial expressions.
• Optimized the internal architecture through experimental analysis to determine the best configuration.
• Utilized the Viola-Jones (VJ) face detector for initial face processing.
• Benchmarked the proposed model against state-of-the-art techniques on FER-2013, CK+, and KDEF datasets.

Main Results:

• The proposed CNN model demonstrates reliable performance in classifying emotion type and intensity.
• Subjective and objective performance analyses confirm the model's effectiveness.
• The optimized architecture mitigates performance degradation issues associated with layer selection in CNNs.
• The model achieved competitive results when benchmarked against existing state-of-the-art methods.

Conclusions:

• The developed DL technique offers an efficient and reliable solution for facial emotion recognition.
• The improved CNN architecture provides a robust foundation for accurate emotion classification.
• Findings have practical applications, particularly for law-enforcement agencies in smart cities.