Improved Real-Time Facial Expression Recognition Based on a Novel Balanced and Symmetric Local Gradient Coding.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Facial Expression Recognition (FER) systems traditionally use local texture coding methods for efficiency and robustness.
• Existing methods require improved accuracy while maintaining real-time performance and low computational load.

Purpose of the Study:

• To propose a novel feature-based FER system using a new local texture coding operator, central symmetric local gradient coding (CS-LGC).
• To enhance the performance of real-time FER systems by improving accuracy and generalization capabilities.

Main Methods:

• A novel operator, CS-LGC, computes gradients in a center-symmetric manner on 5x5 grids, averaging them to reduce noise sensitivity.
• Extracted features are transformed into an eigen-space via Principal Component Analysis (PCA) for efficient representation.
• An Extreme Learning Machine (ELM) is trained for class estimation.

Main Results:

• The CS-LGC operator generates symmetric features, enhancing generalization compared to existing Local Gradient Coding (LGC) variants.
• The proposed system achieved a 95.24% recognition rate on the JAFFE database and 98.33% on the CK+ database.
• The system demonstrates superior performance over existing local texture coding methods.

Conclusions:

• The proposed CS-LGC based FER system effectively balances accuracy, real-time processing, and low computational complexity.
• CS-LGC offers improved feature representation and generalization, making it advantageous for practical FER applications.