Active AU Based Patch Weighting for Facial Expression Recognition.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Machine Learning

Background:

• Facial expression recognition is crucial for human-computer interaction.
• Existing methods often overlook the specific nuances of different expression variations.
• Advanced feature extraction and selection are needed to capture subtle expression details.

Purpose of the Study:

• To develop a novel approach for multiclass facial expression recognition.
• To address the limitations in capturing the specificity of expression variations.
• To improve the generalization and cross-database performance of facial expression recognition systems.

Main Methods:

• The study converts multiclass expression recognition into a triplet-wise recognition problem.
• A new feature optimization model is proposed, incorporating action unit (AU) weighting and patch weight optimization.
• Sparse representation is utilized to detect active AUs for enhanced generalization.

Main Results:

• The proposed algorithm achieved high accuracies of 89.67% on the Jaffe database and 94.09% on the Cohn-Kanade (CK+) database.
• Demonstrated improved cross-database performance compared to existing methods.
• The action unit (AU) weighting and patch optimization effectively capture expression specificity.

Conclusions:

• The triplet-wise recognition framework combined with AU weighting and sparse representation offers a robust solution for facial expression recognition.
• The method significantly improves accuracy and generalization across different datasets.
• This work advances the field of human-computer interaction through more precise facial expression analysis.