Actor-critic guided CDBN with GAN augmentation for robust facial emotion recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Facial emotion recognition (FER) faces challenges due to limited data, noise, and occlusions.
• Existing models often struggle with data diversity and robust feature extraction.

Purpose of the Study:

• To introduce an Actor-Critic Convolutional Deep Belief Network (ACCDBN) for improved FER.
• To enhance FER system accuracy and robustness through a unified generative and reinforcement learning approach.

Main Methods:

• Utilized Conditional Generative Adversarial Networks (cGANs) for data augmentation, expanding minority emotion classes.
• Employed Convolutional Deep Belief Networks (CDBN) for hierarchical texture feature extraction.
• Integrated an Actor-Critic module for reinforcement-driven optimization, refining representations based on prediction accuracy.

Main Results:

• The ACCDBN model achieved 90.4% accuracy and 0.69 MCC on a cGAN-generated dataset via 5-fold cross-validation.
• Demonstrated superior performance compared to baseline models like CNN, LSTM, and ResNet-50.
• Maintained strong performance under noisy and occluded conditions, indicating enhanced robustness.

Conclusions:

• Reinforcement-guided generative learning significantly improves FER accuracy and robustness.
• The proposed ACCDBN offers a promising approach for advanced facial emotion recognition systems.
• The study highlights the effectiveness of combining deep probabilistic learning with reinforcement techniques for complex AI tasks.