Dual-ATME: Dual-Branch Attention Network for Micro-Expression Recognition

Area of Science:

• Computer Science
• Artificial Intelligence
• Biometrics

Background:

• Micro-expression recognition (MER) is challenging due to subtle, rapid facial movements.
• Deep learning with attention mechanisms shows promise but struggles with limited data and feature representation.
• Existing methods often fail to capture the full spectrum of micro-expression (ME) dynamics.

Purpose of the Study:

• To propose a novel Dual-branch Attention Network (Dual-ATME) for enhanced micro-expression recognition.
• To address the limitations of single-scale features in representing micro-expressions.
• To improve the discriminative power of features for more accurate MER.

Main Methods:

• Introduced the Dual-branch Attention Network (Dual-ATME) comprising Hand-crafted Attention Region Selection (HARS) and Automated Attention Region Selection (AARS).
• HARS utilizes prior knowledge for manual feature extraction from regions of interest.
• AARS employs attention mechanisms for automatic extraction of hidden data features.
• Feature fusion combines dual-scale features for effective ME representation learning.

Main Results:

• Dual-ATME demonstrated superior or competitive performance compared to state-of-the-art MER methods.
• Experiments were conducted on spontaneous micro-expression datasets: CASME II, SAMM, and SMIC, including the composite MEGC2019-CD dataset.
• The proposed method effectively learns micro-expression representations through fused dual-scale features.

Conclusions:

• The Dual-branch Attention Network (Dual-ATME) offers a significant advancement in micro-expression recognition.
• Combining hand-crafted and automated attention mechanisms leads to more robust and discriminative feature representations.
• Dual-ATME provides a promising solution for the challenging task of micro-expression analysis.