A Cross-Modal Network for Facial Expression Recognition

Area of Science:

• Computer Science
• Artificial Intelligence
• Biomedical Engineering

Background:

• Deep neural networks are common for facial expression recognition but often miss crucial face properties.
• Existing methods may rely on hierarchical information, limiting their ability to capture holistic facial cues.

Purpose of the Study:

• To propose a novel cross-modal network (CMNet) for enhanced facial expression recognition.
• To leverage biological and structural information, focusing on face symmetry for feature extraction.

Main Methods:

• CMNet learns expression information from whole, left, and right half faces using symmetry.
• A salient facial information refinement module enhances classifier stability.
• A half-face alignment optimization mechanism ensures balanced feature learning.

Main Results:

• CMNet demonstrated superior performance compared to state-of-the-art methods like SCN and LAENet-SA.
• The proposed network effectively extracts complementary facial features through symmetry analysis.
• Experimental validation confirmed the robustness and accuracy of CMNet.

Conclusions:

• CMNet offers a significant advancement in facial expression recognition by integrating structural and biological information.
• The network's ability to learn from face symmetry and refine salient features improves recognition stability and accuracy.
• The findings suggest a promising direction for developing more sophisticated emotion recognition systems.