LEARNet: Dynamic Imaging Network for Micro Expression Recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• Micro-expressions are involuntary, short-lived facial movements reflecting genuine emotions.
• Their subtle nature makes them difficult to perceive and interpret accurately.
• Existing methods struggle to capture the fine-grained details of micro-expressions.

Purpose of the Study:

• To propose a dynamic representation for micro-expressions to preserve crucial facial movement information.
• To introduce a novel Lateral Accretive Hybrid Network (LEARNet) for capturing micro-level facial expression features.
• To enhance the accuracy of micro-expression recognition.

Main Methods:

• Developed a dynamic representation to consolidate video frame information.
• Proposed the Lateral Accretive Hybrid Network (LEARNet) incorporating accretion layers (AL).
• LEARNet utilizes laterally connected convolution layers and cross-decoupled relationships to preserve subtle muscle movement information.

Main Results:

• LEARNet effectively preserves both high-level and micro-level facial expression features.
• Demonstrated significant performance improvements over ResNet on four benchmark datasets (CASME-I, CASME-II, CAS(ME)'2, SMIC).
• Achieved accuracy gains of 4.03%, 1.90%, 1.79%, and 2.82% respectively.

Conclusions:

• The proposed dynamic representation and LEARNet are effective for micro-expression analysis.
• LEARNet significantly advances the state-of-the-art in micro-expression recognition.
• This approach offers a promising direction for accurate emotion detection through subtle facial cues.