LDR-Net: Landmark-Guided Diverse Regional Representation Learning for Facial Expression Recognition

Area of Science:

• Computer Vision
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Facial Expression Recognition (FER) is crucial for human-computer interaction.
• Existing FER methods struggle with real-world variations due to limited spatial structure analysis.
• Overlooking regional features and spatial details hinders representation capabilities.

Purpose of the Study:

• To propose a novel Landmark-guided Diverse Regional Representation Network (LDR-Net) for robust FER.
• To enhance the representation capability of FER systems by preserving structural details and focusing on expression-critical areas.
• To improve FER performance under real-world variations like occlusions and pose changes.

Main Methods:

• Developed a Diverse Regional Feature Extraction (DRFE) module using landmark-guided cropping and cross-level feature integration.
• Introduced a Diverse Representation Learning (DRL) module with a dual-stream mechanism (Transformers and attention) for local and global feature learning.
• Proposed a Hybrid Feature Fusion (HFF) module for hierarchical, joint decision optimization.

Main Results:

• LDR-Net demonstrated superior performance over state-of-the-art methods on multiple FER benchmarks (RAF-DB, AffectNet, SFEW).
• The network showed significant effectiveness on occlusion and pose test sets.
• Cross-scene validation on KMU-FED confirmed LDR-Net's real-world driving applicability.

Conclusions:

• LDR-Net effectively addresses limitations in current FER methods by incorporating dynamic, landmark-guided regional analysis.
• The proposed network achieves robust and accurate facial expression recognition across diverse and challenging scenarios.
• LDR-Net offers a promising advancement for human-computer interaction systems requiring reliable emotion understanding.