Toward Bridging Microexpressions From Different Domains.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Microexpression recognition is challenging due to subtle facial movements.
• Existing methods often fail in cross-domain scenarios due to feature distribution differences.
• Evaluating methods on single databases limits generalizability.

Purpose of the Study:

• To address the challenge of cross-domain microexpression recognition.
• To propose a method that effectively bridges the feature distribution gap between different microexpression databases.
• To improve the robustness and generalizability of microexpression recognition systems.

Main Methods:

• Proposed a novel method combining an auxiliary set selection model (ASSM) and a transductive transfer regression model (TTRM).
• ASSM automatically selects optimal target domain samples as an auxiliary set.
• TTRM learns a joint regression model using source domain samples and the selected auxiliary set to minimize domain discrepancy.

Main Results:

• The proposed TTRM + ASSM method demonstrated superior performance in cross-domain microexpression recognition experiments.
• Experiments were conducted on the SMIC and CASME II databases, validating the method's effectiveness.
• The approach outperformed recent state-of-the-art domain adaptation methods.

Conclusions:

• The TTRM + ASSM method is effective for cross-domain microexpression recognition.
• The proposed approach successfully bridges the feature distribution gap between different microexpression domains.
• This work offers a promising solution for real-world applications requiring robust microexpression analysis across diverse datasets.