Domain Regeneration for Cross-Database Micro-Expression Recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Micro-expression recognition is crucial for applications like lie detection.
• Cross-database recognition faces challenges due to inconsistent feature distributions.
• Existing methods show performance degradation in cross-database settings.

Purpose of the Study:

• To address the performance drop in cross-database micro-expression recognition.
• To propose a novel framework that harmonizes feature distributions across different micro-expression databases.
• To enable effective transfer learning from labeled source data to unlabeled target data.

Main Methods:

• Introduced a Domain Regeneration (DR) framework.
• Developed a domain regenerator to align feature distributions of source and target micro-expression samples.
• Utilized the regenerated samples for classifier training and prediction.

Main Results:

• The DR framework demonstrated improved performance in cross-database micro-expression recognition.
• Experiments were conducted using SMIC and CASME II databases.
• DR outperformed recent state-of-the-art cross-database emotion recognition methods.

Conclusions:

• The proposed DR framework is effective for cross-database micro-expression recognition.
• Domain regeneration successfully mitigates the issue of inconsistent feature distributions.
• DR offers a promising solution for real-world applications requiring robust micro-expression analysis across diverse datasets.