Neural network-based ensemble approach for multi-view facial expression recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Facial expression recognition (FER) is crucial for human-computer interaction.
• Recognizing expressions across diverse poses remains a significant challenge.
• Existing methods often struggle with variations in head orientation.

Purpose of the Study:

• To develop an accurate pose-aware facial expression recognition technique.
• To enhance the performance of ensemble models for multi-pose expression classification.
• To address limitations of current state-of-the-art FER systems.

Main Methods:

• Developed a three-level stacking ensemble model: base-level (binary neural networks), meta-level (pool of binary neural networks), and predictor (Naive Bayes classifier).
• Utilized K-nearest neighbor for pose detection and Voila-Jones face detector for facial region identification.
• Trained and tested the model on the Radboud Faces database using Eigen features.

Main Results:

• Achieved 90% accuracy in pose-aware facial expression recognition.
• Demonstrated superior performance compared to existing state-of-the-art techniques.
• The proposed ensemble approach effectively handles multi-pose facial expression variations.

Conclusions:

• The developed pose-aware FER technique offers significant improvements in accuracy and robustness.
• The three-level stacking ensemble model is effective for classifying facial expressions under varying poses.
• This method provides a promising advancement for applications requiring reliable facial expression analysis.