Automatic Facial Expression Recognition System Using Deep Network-Based Data Fusion.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Facial expression recognition (FER) is crucial for human-computer interaction.
• Existing FER systems often face challenges in accuracy and computational efficiency.

Purpose of the Study:

• To develop a novel and efficient automatic facial expressions recognition system (AFERS).
• To enhance the accuracy and performance of facial expression recognition through a deep network framework.

Main Methods:

• Geometric features extraction.
• Regional Local Binary Pattern (LBP) features extraction.
• Feature fusion using autoencoders and classification with a Kohonen Self-Organizing Map (SOM)-based classifier.

Main Results:

• Achieved high recognition accuracies: 97.55% on the MMI database and 98.95% on the extended Cohn-Kanade (CK+) database.
• Demonstrated superior performance of fused features over individual features or direct concatenation.
• The proposed AFERS showed improved computational efficiency compared to existing methods.

Conclusions:

• The novel deep network framework with autoencoder-based feature fusion and an improved SOM classifier offers a computationally efficient and accurate AFERS.
• Fusion of geometric and LBP features provides a richer representation for facial expressions.
• The proposed system significantly outperforms existing approaches in facial expression recognition accuracy and efficiency.